OJ

'<
G)
CD
CO
o
~
=".
s:
(")

~
D)
::J
 Notice
The information presented in this publication is for the
general education of the reader. Because neither the author
ISA- The Instrumentation, Systems and Automation Society nor the publisher has any control over the use of the
67 Alexander Drive information by the reader, both the author and the publisher
P.O. Box 12277 disclaim any and all liability of any kind arising out of such
Research Triangle Park, NC 27709 use. The reader is expected to exercise sound professional
judgment in using any of the information presented in a
particular application.
Additionally, neither the author nor the publisher have
Printed in the United States of America. investigated or considered the effect of any patents on the
10 9 8 7 6 5 4 3 ability of the reader to use any of the information in a
particular application. The reader is responsible for reviewing
any possible patents that may affect any particular use of the
information presented.
No part of this work may be reproduced, stored in a retrieval Any references to commercial products in the work are
system, or transmitted in any form or by any means, cited as examples only. Neither the author nor the publisher
electronic, mechanical, photocopying, recording or otherwise, endorses any referenced commercial product. Any trademarks
without the prior written permission of the publisher. or tradenames referenced belong to the respective owner of
the mark or name. Neither the author nor the publisher makes
Library of Congress Cataloging-in-Publication Data any representation regarding the availability of any referenced
in Process commercial product at any time. The manufacturer's
instructions on use of any commercial product must be
McMillan, Gregory K., 1946- followed at all times, even if in conflict with the information
Pocket guide to good tuning / by Gregory K. McMillan. in this publication.
p.cm.
ISBN 1-55617-940-5
1. Automatic control-Handbooks, manuals, etc. I. Title .

.ISA wishes to acknowledge the cooperation of those

manufacturers, suppliers, and publishers who granted
permission to reproduce material herein. The Society regrets
any omission of credit that may have occurred and will make
such corrections in future editions.
 Table of Contents

Chapter 1.0-Best of the Basics 1

1.1 Introduction 1
1.2 Actions Speak Louder than
Words 2
1.3 Controller a la Mode 6
1.4 Is That Your Final Response? 7
1.5 The Key to Happiness 30
1.6 Nothing Ventured, Nothing
Gained 34
1.7 Process Control as Taught
versus as Practiced 42

Chapter 2.0- Tuning Settings and Methods 43

2.1 First Ask the Operator 43
2.2 Default and Typical Settings 43
2.3 The General-purpose
Closed-loop Tuning Method 47
2.4 The Shortcut Open-loop
Method 53
2.5 Simplified Lambda Tuning 58
2.6 Set Point Response and Load
Rejection Capability 62

Chapter 3.0-Measurements and Valves 63

3.1 Watch Out for Bad Actors 63

1.<
 3.2 Deadly Dead Band 64 6.6 Crystallizers 130
3.3 Sticky Situations 74 6.7 Distillation Columns 132
3.4 Fouled Sensors 78 6.8 Dryers 135
3.5 Noisy Measurements 81 6.9 Evaporators 137
6.10 Extruders 138
Chapter 4.O-Control Considerations 85
6.11 Fermentors '" 140
4.1 Auto Tuners 85 6.12 Heat Exchangers 142
4.2 Uncommonly Good Practices 6.13 Neutralizers 143
for Common Loops 85
6. 14 Reactors.................................. 145
4.3 Dead-Time Compensation and
6.15 Remote Cascade 146
Warp Drive 88
6.16 Sheets and Webs 147
4.4 I Have So Much Feedforward,
I Eat before I Am Hungry 92 Chapter 7.0-Adaptive Control 151
4.5 Cascade Control Tuning 95
7.1 Learning the Terrain................. 151
4.6 Keep the Secondary Loop
on the Move 98 7.2 Watching but Not Waiting 152
7.3 Shifting into High Gear 154
Chapter 5.0- Troubleshooting 99 7.4 Back to the Future 156
5.1 Patience, Heck, I Need to
Appendix A-Technical Terms in
Solve the Problem 99
Process Control that Are
5.2 Great Expectations and Used Interchangeably 159
Practical Limitations 107
Appendix B-For Math Lovers Only 161
Chapter 6.0- Tuning Requirements for
Various Applications 113 Appendix C-An Integral Part of Tuning 165
6.1 Batch Control 115
6.2 Blending 118 Appendix D-Closed Loop Time Constant.. 169
6.3 Boilers 119
6.4 Coils and Jackets 125
6.5 Compressors 128
I.O-Best of the Basics

1.1 Introduction
Welcome to the wonderful world of proportional-
integral-derivative (PID) controUers. This guide wiD
cover the kry points of good tuning and provide more
than seventy rules of thumb. First, let's blow away
some myths:

Myth 1 - It is always best to use

one controller tuning method.
False; the diversity of processes, con-
trol valves, control algorithms, and
objectives makes this impractical.

Myth 2 - Controller tuning settings

can be computed precisely. Not so.
The variability and nonlinearity in
nearly all processes and control
valves makes this implausible. Any
effort to get much more than one significant
digit is questionable because any match to the
plant is momentary. If you run a test for an auto
tuner or manually compute settings ten times,
you should expect ten different answers.

Roughly 75 percent of process control loops

cause more variability running in the automatic
node than they do in the manual mode. A third troller for a reverse-acting process except as
)f them oscillate as a result of nonlinearities noted later in this guide. Correspondingly, you
uch as valve dead band. Another third oscillate should select reverse control action for a direct-
>ecauseof poor controller tuning. !be .re~ain- acting process so an increase in process variable
ng loops oscillate because of defiClenClesill the measurement will cause a decrease in controller
:ontrol strategy. A well-designed control loop output that is proportional to its gain setting,
~ith proper tuning and a responsive control . except as noted later. A direct-acting process is
ralve can minimize this variability. Because this one in which the direction of the change in the
neans you can operate closer to constraints, process variable is the same as the direction of
~ood tuning can translate into increased produc- the change in the manipulated variable. A
jon and profitability. reverse-acting process is one in which the direc-
tion of the change in the process variable is
1.2 Actions Speak Louder than Words opposite the direction of the change in the
manipulated variable. The manipulated variable
The very first settings that must be right.are the is most frequently the flow through a control
controller and valve actions. If these actlOns are valve, but it can also be the set point of a slave
not right, nothing else matters. The con.troll.er loop for a cascade control system or variable
output will run off scale in the wrong dIrectIon speed drive.
regardless of the tuning settings.
The valve action sets the display. For example, it
The controller action sets the direction of a determines whether a 100 percent output signal
change in controller output fr?m its propor- , corresponds to a wide open or a fully closed
tional mode for every change m the controller s valve. It also determines the direction of a
process variable (feedback meas~rement). If y?U change in the actual signal to the control valve
choose direct action, an increase m process van- when there is a change in the controller's out-
able (PV) measurement will caus~ an incr:ase i~ put. In some analog controllers developed in the
controller output that is propomonal to Its gam 1970s, such as the Fisher AC2, the valve action
setting. Since the controller action must be the affected only the display, not the actual signal.
opposite of process action to ~rovide ~eedback To compensate for this lack of signal reversal for
correction, you should use a duect-actmg con- a reverse-acting valve (i.e., an increase-to-close or

J.,
fail-open valve), the control action had to be the
opposite of the action that would normally be
appropriate based on process action alone. For- Process Valve Signal Controller
Action Action Reversal Action
tunately, the valve action corrects both the dis-
play and the actual valve signal in modern Direct Increase- No Reverse
controllers, so the control action can be based Open
solely on process action. However, the user
should verify this before commissioning any Reverse Increase- No Direct
loops. In control systems that use fieldbus Open
blocks, the valve action should be set in the ana-
Direct Increase- Yes Reverse
log output (AO) block rather than in the PID Close
controller block. This ensures that the "back-cal-
culate" feature is operational for any function Reverse Increase- Yes Direct
blocks (split range, characterization, and signal Close
selection) that are connected between the PID Direct Increase- No Direct
and AO blocks. The signal can also be reversed Close
in the current-to-pneumatic transducer (liP) or
in the positioner for a control valve. Before the Reverse Increase- No Reverse
advent of the smart positioner, it was preferable Close
for the sake of visibility and maintainability that
any reversal be done in the control room rather Which brings us to rule of thumb number one.
than at the valve. It is important to standardize
on the location of the signal reversal to ensure ' Rule 1- The controller action
that it is done and done only once. Table 1 sum- .~.. should be the opposite of the pro-
marizes how the controller action depends upon cess action unless there is an
both the process and valve actions and on the .•• • increase-to-dose (fail-open) con-
~
signal reversal. trol valve for which there is no
reversal of the valve signal. This means that
you sho~ld use reverse and direct-acting control-
lers for dlTect and reverse-acting processes,

.~.
respectively. The valve signal can be reversed for • Man (manual) - The operator manually sets
a fail-open valve at many places, but it is best the output. PID action is suspended (open
done in the AO block of the control system. loop).

1.3 Controller it la Mode • IMan (initialization manual) - PID action

is suspended because of an interruption in
The names for the operational modes of the PID the forward path of the controller output.
vary from manufacturer to manufacturer. Thus, This is typically caused by a downstream
the FOUNDATION'TM Fieldbus modes listed next block that is not in the cascade mode. The
provide a uniformity that can be appreciated by controller output is back-calculated to pro-
all. vide bumpless transfer.
• Auto (automatic) - The operator locally • RCas (remote cascade) - The set point is
sets the set point. PID action is active remotely set, often by another computer.
(closed loop). In older systems, this mode is PID action is active (closed loop). This
also known as the local mode. mode is also known in older systems as the
• Cas (cascade) - The set point comes from supervisory mode.
another loop. PID action is active (closed • ROut (remote output) - The output is
loop). This is also known in older systems remotely set, often by a sequence or by
as the remote mode or remote set point another computer. PID action is suspended
(RSP). (open loop). In older systems, this mode is
• LO (local override) - PID action is sus- known as direct digital control (DDC).
pended. The controller output tracks an
external signal to position the valve. This 1.4 Is That Your Final Response?
mode is typically used for auto tuning or to
coordinate the loop with interlocks. In The contribution that the proportional action
older systems, it is also known as output makes to the controller output is the error multi-
tracking. plied by the gain setting. The contribution made
by the integral action is the integrated error mul-
tiplied by the reset and gain setting. The reset
setting is repeats per minute and is the inverse of
integral time or reset time (~inutes pe~ repeat). When derivative action is used with a time con-
FOUNDA TION1'M Fieldbus w111standardlze the stant there is a built-in filter that is about one-
reset time setting as seconds per repeat and the eighth (118) of the rate setting. However, you
rate (derivative) time setting as seconds. The should use set point velocity limits to prevent a
contribution made by derivative mode is the rate jolt to the output when there is a large step
of change of the error or process variable in change in error from a manually entered set
percent (Ofo), depending upon the trl?e of ~go- point. This is particularly important when you
rithm, multiplied by the rate and gam settmgs. are using large gains or derivative action based
on control error.
When derivative action is on the process vari-
able instead of on the control error, it works The PIO algorithm uses percentage (Ofo) input
against a set point change. (The contr~l error is and output signals rather than engineering units.
the difference between the process vanable and Thus, if you double the scale span of the input
the set point.) The reason for this is that it (error or process variable), you effectively halve
doesn't know the process variable should be the PIO action. Correspondingly, if you double
changing initially and that the brakes.should the scale span of the output (manipulated vari-
only be applied to the process when It . able), you double the PIO action. In fieldbus
approaches set point. Using .derivation aetlon. blocks, both input and output signals can be
that is based on the change m control error Wlll scaled with engineering units. Using output sig-
provide a faster initial takeoff and w~ll~uppr~ss nal scaling will facilitate the manipulation of the
overshoot for a set point change. ThlS ISparticu- slave loop's set point for cascade control. For
larly advantageous for set points driven for batch controllers that use proportional band, you need
control, advanced control, or cascade control. to divide the proportional band into 100 percent
The improvement can translate into shorter to get the equivalent controller gain. Propor-
CYcleor transition times, an enhancement of the tional band is the percentage change in error
ability of slave loops to mitigate upsets, and less needed to cause a 100 percent change in output.
off-spec product because overshoot has been
diminished. Proportional mode is expressed by the following
equation (note that adjustments are gain or pro-
portional band):
 Pn =Kc* En Reset action (repeats/minute) = 60 / T
i
When the derivative mode acts on the process
variable in percent (%PV), the equation becomes Where:
as follows for the series form: = error at scan n (Ofo)

P=l(*D = controller gain (dimensionless)

n "'c n
= controller proportional band (%)
Integral mode is expressed by the following = contribution of the proportional mode
equation (note that adjustments are integral time at scan n (%)
or reset): OfoPVn= process variable at scan n (Ofo)
In = contribution of the integral mode at
In = Kc * 1/Tj * (En* Ts) + In-1 scan n (%)
When the derivative mode acts on %PV, the On = contribution of the derivative mode at
scan n (Ofo)
equation becomes as follows:
= derivative time or rate time setting
In = Kc * l/Tj * (Dn* Ts) + In-1 (seconds)
= integral time or reset time setting
The equation for derivative mode is as follows (seconds/repeat)
(adjustments are derivative time or rate):
= scan time or update time of PID
controller (seconds)

. _ Rule 2 - If you halve the scale

.... ' span of a controlled (process)
. vari~ble or double the span of a
~ mampulated variable (i.e., set
Note the inverse relationships across controllers! point scale or linear valve size),
you need to halv~ the controller gain to get
Gain (Kc) = 1000/0 / PB the same PID actIon. For controlled variables
the PID gain is proportional to the measureme~t
calibration span. Often these spans are narrowed
since accuracy is a percentage of span. For con-
trol valves, the PID gain setting is inversely pro-
portional to the slope of the installed valve
characteristic at your operating point. If a valve
is sized too small or too large, the operating o -
point ends up on the flat portion of the installed u
.~

valve characteristic curve. For butterfly valves, . <I

the curve gets excessively flat below 15 percent II
N
and above 55 percent of valve position. o
u
Figure 1 shows the combined response of the ~
PID controller modes to a step change in the <I
process variable (%PV). The proportional mode
provides a step change in the controller output .•..
(~%COl)' If there is no further change in the .-=o
%PV, there is no additional change in the out- C.
put even though there is a persistent error (off-
set). The size of the offset is inversely
proportional to the controller gain. Integral
action will ramp the output unless the error is
zero. Since the error is hardly ever exactly zero,
reset is always driving the output. The contribu-
tion made by the integral mode will equal the o -
contribution made by the proportional mode in u
the integral time (~%C02 = ~%COl)' Hence, ~
<I
the integral time setting is the time it takes to
repeat the proportional contribution (seconds -==~.-..°
.- -
per repeat). The contribution made by the deriv-
OIl
ative mode for a step change is a hump because 00
of the built-in filter, which is about one-eighth
(1/8) of the rate setting. Otherwise, there would
be a spike in the output.

If the temperature is below set point for a reac-

tor, as shown in Figure 2, should the steam or
water valve be open? After looking at a faceplate
or the digital value for temperature on a graphic
display, most people think the steam valve is
open when the temperature is below set point.
Reset provides a direction of action that is con-
sistent with human expectation. However, the
proper direction for a change in controller out-
put and the split-ranged control valve depends
upon the trajectory of the process variable (PV).
If the temperature is rapidly and sharply increas-
ing, the coolant valve should be opening. Gain
and rate action will recognize that a set point is
being approached and position the valves cor-
rectly to prevent overshoot. In contrast, reset has ";
no sense of direction and sacrifices future results '5 :;1 N
It)
~
for immediate satisfaction. Most reactors, evapo- §~
rators, crystallizers, and columns in the process
industry have too much reset. On two separate
-.
u-"
,g ~
u
o ~I
~I

I I
;
to-

applications in chemical plants, for example, it •"

was reported that the controller was seriously
malfunctioning because the wrong valve was E " ••
c > c: ~ " ..
supposedly open, when in reality it was just gain ,,-
_Cc.
•• >0
" " > c:
'5 - "
~oc.
> 0
and rate doing their job to prevent overshoot.
Figures 3a and 3b show the effect of gain setting
on a set point response. If the gain is too small,
the approach to set point is too slow. If the gain
is too large, the response will develop oscilla-
tions. If the dead time is much larger than the
time constant and the gain action is too large
i~
compared to the reset action, the response will j

momentarily flatten out (falter or hesitate) well i( I

below the set point. Loops that are dead-time .- "' " '.-, ;' ' 1' .
dominant tend to have too much gain and not
enough reset action. For nearly all other loops in ~I '
tl~·_
the process industry, the opposite is true. The
size of the dead time relative to the time con-
stant and the degree of self-regulation (ability to
reach a steady state when the loop is in manual)
determine which tuning methods you should
:.\ . i.
employ.

It is particularly important to maximize the gain

.
--T 't m '.-1r:!-:'

so you can achieve tight control of loops that ............... ~~ .. \........

\
. -- ~
can ramp away from set point, such as gas pres- .•. \
""
, ,

sure. Maximizing the gain is also important for )~"

..
loops that can run away from set point, such as .. '-
', ':-. .
. ,;)

an exothermic reactor temperature. Finally, gain

must be maximized to speed up the set point
response for advanced control, batch control,
and cascade control as well as for startup
sequences. However, gain readily passes variabil-
ity in the process variable to the output, which
can increase overall variability and interaction.
 You must set the process variable filter to ensure
that fluctuations from measurement noise stay
within the dead band of the control valve. The
variability amplification by gain can be particu-
larly troublesome if there is inadequate smooth-
ing because of insufficient volumes or mixing.
These loops tend to be dead-time dominant, and
they exhibit abrupt responses (i.e., no smoothing
by a time constant). Thus, you should reduce
the gain setting for loops on pipelines, de super-
heaters, plug flow reactors, heat exchangers,
static mixers, conveyors, spinning (fibers), and
sheets (webs).

:..
.-. - Rule 3 - Increase the gain to
.-- -.~.... - ..... --';
~.,; •... .' -. achieve tight level, stirred reactor,
\ t~ - • or column control and to speed
II
\ ...
i E :
;'l ~
~ -- up the set point response for
..
------------: _----\
\
_._- - _ ..
.!:E.
advanced, batch, cascade, and
•• ••\
sequence control. For high gains it is especially
\
important that you set a process variable filter to
.\
~?:c- ensure that fluctuations in the PID output are
smaller than the valve dead band. You should
also establish set point velocity limits so the out-
put doesn't jerk when the operator changes the
set point.

BEST OF THE BASICS

 Rule 4 - Decrease the gain to Rule 6 - For residence time con-
provide a smoother, slower, and trol and for material balance con-
more stable response; reduce trol, use as high a gain for level
interaction between loops; and control as possible, such as 5.0,
reduce the amplification of vari- that does not wear out the valves
ability. Pulp and paper, fiber, and sheet pro- or upset other loops. Continuous reactors,
cesses use less gain (higher proportional band) evaporators, crystallizers, and thermosyphon
and more reset action (smaller reset time). They reboilers often need to maintain a level accu-
also are much more sensitive to variability from rately, particularly if they are being pushed
valve dead band. The dead time is larger than beyond nameplate production rates. Level con-
the process time constant for most of these trol also needs to be tight when it manipulates a
loops. reflux flow for a column or a reactant makeup
flow for a recycle tank.
~ Rule 5 - For level control of
The Ziegler-Nichols O. G. Ziegler and N. B.
Nichols, "Optimum Settings for Automatic
~ ~~:~:~~~~~~~:~e~~to Controllers," Transactions if the ASME, vol. 64,
help prevent the tank from over- Nov. 1942, p. 759) tuning method depicts a goal
flowing or running dry. You can estimate the of quarter-amplitude response where each subse-
gain as follows: it is the maxiI?~m needed . quent peak in an oscillatory response of the pro
change in controller output dIvIded by the dif- controller is one-fourth of the previous peak.
ference between the high- and low-level alarms. This goal will generally provide a minimum peak
For example, if a 60 percent change in controller error, but the controller is too close to the edge
output will always keep the level between 30 per- of instability. An increase of just 25 percent in
cent and 70 percent, the correct controlle~ gain the process gain or dead time can cause severe
is 1.5. You can add an error-squared algonthm, oscillations. By simply cutting the controller
where gain is proportional to error, to attenuate gain in half, you will make Ziegler-Nichols tun-
control action near the set point. This is a gener- ing sufficiently robust. You will also make it
ally effective way to reduce the effect of noise for look more like Internal Model Control (IMC)
all types of level control loops on the manipu- tuning for stirred reactors, evaporators, crystal-
lated flow.
lizers, and columns. If the controller manipu-
lates a control valve instead of a flow set point,
you need to reduce the gain for the steepest part
of the installed valve characteristic curve.

For loops that will reach a steady state when put

in manual, decreasing the gain will always
improve their stability. For exothermic reactors
where a runaway condition can develop, there is
a window of allowable gains in which instability
can be caused by a gain that is too small or too
large. For level loops with reset action, there is
also a window in which a gain that is too small
can cause slow, nearly sustained oscillations. i i
" " I
~
.
"~ "~
~.' i
:
!
... ... Rule 7 - If you have a quarter- ( .~!
;i: ~
.........
.... i', amplitude response, yon should ~n
t ~1:. ~
. decrease the gain to promote sta-
bility. Generally, halving the con-
., 1) ~I;
C E C

.!
- ..•
~
~.
~

~ troller gain is sufficient. This

I
··t··· t!
::
(
assumes that the reset action is not excessive \
\ .....
.•.... ;,
(i.e., integral or reset time is too small). If the ;""::£:~ti-~------ ~
!!
manipulated variable is a control valve and the "

operating point moves from the flat to the steep

portion of the installed characteristic, you may
need to cut the gain by a factor of three or four.

Figure 4 shows the effect of reset setting on a set

point response. If the reset action is too small
(i.e., the reset time too large), it takes a long time
to eliminate the remaining error or offset. Rule 8 - If there is excessive
Whereas insufficient gain slows down the initial overshoot and oscillation,
approach, inadequate reset action slows down decrease the reset action (increase
the final approach to set point. If the reset action the reset time). If the current reset
is too large (reset time is too small), there is time setting is less than half the
excessive overshoot. The addition of reset action oscillation period for loops on mixed volumes
reduces controller stability and is dangerous for or less than a tenth of the period for any loop,
exothermic reactor temperature loops. Polymer- there is too much reset action. If the oscillation
ization reactors often use proportional-plus- period changes by 25 percent or more as you
derivative temperature controllers (i.e., no inte- change reset action, it is a sign that the oscilla-
gral action). For level loops, using a reset time of tion is caused or aggravated by reset action.
smaller than 3,000 seconds per repeat makes it
necessary to use relatively high gains (5.0 for i'; . Rule 9 - For leve1loops with low
large volumes) to prevent slow, nearly sustained ...........>
controller gains « 5.0), reset
oscillations. action must be greatly restricted
~
' · •• (i.e.• reset time increased to 50 or
Both reset and gain can cause oscillations, and a more minutes) to help mitigate
high gain setting can aggravate an overshoot nearly sustained oscillations. Level loops are
caused by reset action. For this reason, checking the opposite of most other loops in that you can
the ratio of the reset time to the oscillation increase the reset action (decrease the reset time)
period can help you distinguish the main culprit. as you increase the gain (see page 109 for more
You will need to decrease the reset action (i.e., detail).
increase the reset time) of the loop if the ratio is
much less than 0.5 for vessel or column tempera- Figure 5 shows the effect of rate setting on a set
ture, level, and gas pressure control or if it is point response. A rate setting that is too small
much less than 0.1 for flow, liquid pressure, increases any overshoot caused by reset action.
pipeline, spin-line, conveyor, or sheet (web) A setting that is too large causes the approach to
control. set point to staircase and if large enough it can
cause fast oscillations.
 Rate can be used whenever there is a smooth
and slow open-loop process response. Rate can-
not be used where there is excessive noise, chat-
ter, inverse response, or an abrupt response. It
can be used primarily on temperature loops that
manipulate heating or cooling where there is a
temperature transmitter with a narrow calibra-
tion span (large spans cause analog/digital [ND]
chatter). Analyzers that have a cycle time (sam-
ple and hold) have signals that are too noisy and
abrupt for derivative action. Continuous con-
centration measurements made by electrodes or
inline devices (such as capacitance, conductivity,
density, microwave, mass spectrometry, pH, and
viscosity for mixed volumes) may allow you to
use derivative mode if you add a filter to attenu-
ate the high-frequency measurement noise.

If the measurement is continuous but is used for

pipeline control, the control response is too
abrupt. This is because the loop dead time is
large compared to the largest time constant.

There is noise in every process variable. If the

noise is greater than the measurement resolution
limit, the loop sees it. If it causes an output
change greater than the final element or control
valve resolution limit, loop variability will be
worse in auto than in manual, particularly if
there is any rate action. While it is true that
increasing the measurement and valve resolution plete trajectory and not the current measure-
limit reduces the loop's sensitivity to noise, it ment like the PIo controller. Poor valve
also degrades its ability to recover from load resolution is as much a problem for the MPC as
upsets. If the measurement doesn't see or correct
for the PIo, unless you set the minimum size of
an upset because the control error is smaller
a move to exceed the resolution limit. The reso-
than a resolution limit, the control loop perfor-
lution limit is unfortunately a function of direc-
mance deteriorates. tion and position.
Poor control valve resolution is a problem for
any loop. If the valve doesn't move, there is no ~............. . ?"JZ.................. . [l.ule 10- Rate is primarily used
.i( loops that have In temperature
correction. On the other hand, poor measure- .. narrow span transmitters and slow
ment resolution is primarily a problem for PIo ... .. scan times. However, it is also used
temperature loops. The large-span ranges found for continuous analytical measure-
in computer input cards or transmitter calibra- ments such as capacitance, conductivity, den-
tion can result in a resolution limit of 0.2S0F or sity, mass spectrometry, microwave, pH, and
more. It is impractical to expect control that is viscosity for concentration control of mixed vol-
tighter than twice the resolution limit or O.SOE umes. If the temperature control is not intended
Also, for a change of 0.2S0F per minute, an addi- for a mixed volume, the rate setting is very small
tional minute of dead time is introduced into (6 seconds) and is mostly used to compensate
the loop. Short scan times also introduce chatter for the thermowell lag. In every case, you must
because the actual temperature change is small set the measurement filter large enough to keep
compared to the noise (i.e., the signal-to-noise output fluctuations within the valve dead band.
ratio is too small). The scan time for most tem-
perature loops is too short. Poor measurement Rule 11- Most temperature and
resolution and fast scan times are the main rea- .•
level loops chatter because the
sons why rate cannot be used in temperature PID scan time is too small. The
loops. Model Predictive Controllers (MPC) are ~ • true temperature or level change
much less sensitive to measurement resolution within the scan time must be large
because the scan time is large and the move in compared to the NO resolution limit (0.050/0);
the manipulated variable is based on the com- otherwise, the signal-to-noise ratio is too small.
 other time constants are undesirable and create
1.5 The Key to Happiness additional dead time. The major time constant
When I was four years old and sitti~g on my can be approximated as follows: it is the time it
daddy's knee, he said, "So~, I ~ave Jus.tone takes to reach about 63 percent of the final value
thing to say to you-dead t:J.m~.~ell, It took ~e after the dead time. It takes one dead time and
forty years to appreciate the significance ?f his four time constants for a response to reach 98
words of wisdom. If a loop has no dead ~Ime?r percent of its final value.
noise, perfect control is possible ~nd.an mfimte-
gain permissible. There ISno turun~ Issue. With- The open-loop steady-state gain is the percent-
out dead time, I would be out of a Job. age change in the process variable for a percent-
age change in output after all transients have
Dead time is that period of time from the start died out. The controller is in manual. A high
of a disturbance until the controller makes a cor- steady-state gain is both a curse and blessing. It
rection that arrives at the same point in the loop can improve the control of the true process vari-
at which the disturbance entered. The controller able by making the measurement more sensitive.
needs to see the upset, react to it, and ~et the This is particularly important for inferring com-
correction to the right place. To appreciate dead position from temperature or concentration
time imagine that you go to a party and start from pH. However, a high steady-state gain can
drinking. The period of time between the first make it more difficult-and sometimes impossi-
drink and when you eventually bypass the next ble-to control the measured variable. For exam-
round is dead time. ple, excessive oscillation can result from a small
amount of stick and slip in a reflux valve for an
While zero dead time is not possible, ~ dec~e~se acrylonitrile and water distillation or in a reagent
in dead time reduces the effect of nonlmeantles, valve for sulfuric acid and sodium hydroxide
such as changes in the time constants and neutralization. A low steady-state gain often
steady-state gains of t~e loop, and makes the results when a throttle position is on the exces-
loop easier to tune. It I~deSirable that the larg~st sively flat portion of the installed valve charac-
loop time constant b~ m the process because It teristic. The process variable will wander and
will slow down the divergence of the measure- respond to the conditions of a related loop more
ment from the set point during an upset and than from the manipulated variable.
give the controller a chance to catch up. All

BEST OF THE BASICS

 The block diagram in Figure 6 shows all the dead
times, time constants, and steady-state gains in a
control loop. Feedback control corresponds to
the signal making one complete circle around
the block diagram. The total dead time is

I
I approximately the sum of all the pure dead
.\ times and small time constants as you traverse
I
I

I cl
Q) ,
the loop. The dead time from valve dead band is
inversely proportional to the rate of change of
e
~
...-
~I
III

~
I

!
the controller output, and the dead time from
transportation delays is inversely proportional to
ca ~! throughput. The diagram in Figure 6 has many
Q
~ I uses-including preventing guests from overstay-
-=
e
Co
>
Q
~ I!
I
I
ing their welcome. Just show this slide and start
talking about dead time and you will be amazed
.se I
I
at how quickly the place empties. There is a
safety issue, however, in that guests can get tram-
I I
!
..c ~c: I
I
pled in the rush for the door.
U Co? I
I
•••
.•
I
~ ~III ... I
I
I The following list summarizes the three key vari-
Woe 0.. Q) ~ i

::> >
CO ...
(5 I
I
ables and their relationship to loop performance:
"0
al, > 'E
0
I
I
()
51 I
0
U
I
I
I 1. Dead Time or TimeDelay {'tcJ
I
I ~ II
I
I
! u
I The most important of the three key loop
:II: II
variables
0
ii:
Delays controller's ability to see and react
to upset
Perfect control is possible for zero dead
tIme
 Nonlinearities become less important as the sented in this section, you should skip ahead to
dead time decreases Section 2.0 on Tuning Settings and Methods.

2. Time Constant ('t) The controller gain is inversely proportional to

the open-loop gain for all types of loops. For
Better control is possible for a large process
loops that are dead-time dominant (i.e., dead
time constant downstream of the load
time is much greater than the largest time con-
upsets stant), the controller gain can be simply set as
Time constants in series create dead time one-fourth of the inverse of the open-loop gain.
For these same loops the reset time can be set as
Measurement time constants give the illu-
one-fourth of the dead time. The result is a
sion of providing better control
smooth stable response similar to what you
would get from Lambda tuning, which will be
3. Steady-State Gain (K) discussed (see Section 2.5) as one of the pre-
The valve and process steady-state gains are ferred tuning methods for dead-time dominant
usually nonlinear loops. For loops with a healthy time constant,
the controller gain is also proportional to the
High steady-state gain causes overreaction
ratio of the time constant to the dead time. For
and oscillations
these loops, the smooth response that is pro-
Low steady-state gain causes loss of sensitiv- vided by a large time constant (i.e., residence
ity and wandering time) of a mixed volume enables you to use a
higher controller gain for tighter concentration,
1.6 Nothing Ventured, Nothing gas pressure, and temperature control.
Gained
If the loop time constant is much larger than the
There is a lot to be gained from having a better dead time ('t/'td » 1), then the controller gain is
understanding of the overall open-loop gain proportional to this ratio besides the inverse of
(Ka). However, if you don't have time ~o study the open loop again:
the relationships revealed by the equatIOns pre-
I\: == 0.25 * 't / ('td * Ka)
If the loop is dead-time dominant (tl'td « 1), and cancels out the flow ratio gain. However, for
then the controller gain is proportional to just loops on agitated vessels and columns the con-
the inverse of the open loop gain: troller gain is proportional to the process time
constant (residence time), which is inversely pro-
I\: = 0.25 * 1/ Kc portional to feed flow. Since, as we mentioned
previously, the controller gain is also inversely
The manipulated variable gain is linear for.a proportional to the open-loop gain, the intro-
variable speed drive (VSD). For a control valve it duction of an equal-percentage valve characteris-
is the slope of the installed characteristic curve tic is bad news because the flow ratio gain was
and is typically very nonlinear. For operating cancelled out by the process time constant for
points on the upper part of the installed charac- the controller gain.
teristic curve of a butterfly valve, the slope can
be so flat that the valve gain approaches zero. The process gain for flow loops is 1. For pressure
loops downstream of a compressor, fan, or
Note that for concentration and pressure loops, pump, the process gain is the slope at the operat-
the process variable should be plotted against a ing point on the characteristic curve of the com-
flow ratio (e.g., column temperature versus distil- pressor, fan, or pump. At low flows, the slope
late-to-feed ratio, heat exchanger temperature can be so flat that the process gain approaches
versus coolant-to-feed ratio, and pH versus zero.
reagent-to-feed ratio). The process gain is the
slope of this curve and is highly nonlinear. At The controlled variable gain is the inverse of the
high flow ratios, the slope can be so flat that the calibration span for the controlled variable. It is
process gain approaches zero. The open-loop linear except when a square root extractor is not
gain has a flow ratio gain that is inversely pro- used on a differential head meter.
portional to the feed flow, which is an additional
source of nonlinearity for pipeline, desuper- Rule 12 - For dead-time-domi-
heater, static mixer, and exchanger temperature
or concentration control. In these applications,
£3'!1 nant loops, set the gain equal to
about one-fourth of the inverse of
an equal-percentage valve characteristic is desir- ~ the open-loop gain and the reset
able since the valve gain is proportional to flow time (seconds per repeat) equal to
about one-fourth of the dead time. Rate action
For pressure loops, the open-loop gain is a
is not used. This provides a smooth stable
steady-state gain with a process gain 1), which
response similar to Lambda tuning.
is the slope of the pump or compressor ~urve,
and no flow ratio gain (Kfr omitted and 1),v =

t3 5~§~fik~~~~]i~~
curve slope):

.. . loop more linear. ThIS assumes

that the installed characteristic is close t~ ~e For level loops, the open-loop gain is a ramp rate
inherent equal-percentage trim charactensuc. and u~es 1\ ~nstea? of 1),v to denote an integra-
tor gain. 1\ ISthe Inverse of the product of fluid
For concentration and temperature loops, the density and the vessel's cross-sectional area.
open-loop gain is a ste~dy~state~ain that has a There is no flow ratio gain (Krr omitted):
flow ratio gain 4 whIch ISthe Inverse of the
feed flow:

~CV ~MV ~FR ~PV ~CV K· = - 1 _

Ko = ~co = ~CO· ~MV· ~FR • ~PV 1 density. area

Kc = Kmv * Kt-r * Kpv * Kcv

Ff = feed flow (Pph)
1
Kfr =F 1\ =level loop integrator gain (ftJlb)
f
~ =overall open-loop gain (l/hr for level,
For flow loops, the open-loop gain i~ a steady- otherwise dimensionless)
state gain that has a unity process gam and no I<mv = manipulated variable (valve or VSD)
flow ratio gain (Ktr omitted and ~v = 1): steady-state gain (Pph/OJo)
Krr = flow ratio steady-state gain (inverse of
feed flow) (lIpph)

BEST OF THE BASICS

1
I; Kpv = process variable steady-state gain
(wtfrac/%) (degC/%) (psi/%)
Kcv = controlled variable steady-state gain
(%/wtfrac) (0/0/ degC)(%/psi)
(%/pph)(%/ft) c
c
~ Cl.
;>
~CO = change in the controller output (%) ~
=-
~cv = change in the controlled variable (Ofo) ~

-c2- -
II
<J
~PV = change in the process variable (wtfrac)
(degC) (psi) (Pph) (ft) u <J
•..•. 0
~FR = change in the flow ratio (Pph/pph) ;> U
c. =- ~ >
0

~MV = manipulated variable or valve flow ~

<J
<J Cl.
~
(Pph) ~ II
...=
<l
~
=
-; ~- ~
0
"'l --5 -
Eo-
Figure 7 shows the dead time, time constant, ~ IC

-
••• ClO
and steady-state gain for a process that reaches a Q ;> Eo:'
steady state (i.e., self-regulating) and for an inte- =-
grator such as level. In both cases, the process
variable response is to change the controller out-
-~
oS
I c
Cl.
~
put with the loop in manual (open-loop to- ;> 0
u =- U
response). Figure 7 also shows the equations ~ ~~ ~
c

used to approximate an integrator gain as equal -."-

~ <J ~
<J
ellSf'-l
Q
..:l~
Cl.
to the initial ramp rate of a self-regulating II
c
response. This initial ramp rate is approximately ~
the steady-state gain divided by the dominant
time constant. Auto tuners that compute set- ~
f'-l-
f'-l,Q
tings from an open-loop response use this ~ eIIS-
("j
Q --I. ~c
approximation to convert an integrator into an I. eIIS_
=-;>
equivalent
time constant and steady-state gain because they
cannot handle integrator gain. This method 2.0- Tuning Settings and
works well when the dead time is small com-
pared to the ramp rate. The dead time is the
Methods
time required for the PV to get out of the noise
band.
2.1 First Ask the Operator
1.7 Process Control as Taught versus
You should test new.tuning settings by changing
as Practiced t~e c~ntroller set POInt and output in both
Most books, courses, and papers concentrate on dIrectIOns and observing whether the response is
a set point response and if a load is introduced, smooth and fast enough for these set point
it is usually shown as an addition to the process changes and load upsets. Don't leave these new
output, which is really just measurement noise. s~ttings over night until you are sure they are
In the process industry, the real load upsets, ~Ight. Before entering new settings or trying tun-
such as feed flow, temperature, and composi- mg methods, ask the operator for permission to
tion, enter as inputs to the process as shown in tune the loop. Also ask him or her what the
Figure 6. However, tuning studies that concen- maximum allowable size can be for each direc-
trate solely on the set point response have seri- tion of a step .change in the controller set point
ously compromised the load rejection capability and output WIthout causing a problem. You
for processes with a large process time constant. s~~uld put most analog controllers and some
The optimum load response should reach a peak dIgItal c.ontr,?llers in manual before entering a
in just over one total loop dead time and show a change m gaIn to avoid bumping the controller
return to set point that is just as fast. This is gen- output.
erally achieved by the controller initially over-
driving its output past its final resting position. 2.2 Default and Typical Settings
The load response can easily be checked by
momentarily putting the controller in manual, !ables ~ and 3 show the default and typical tun-
making a step change in the controller output, ~ngsettmgs for ~ontrollers whose reset setting is
and then returning the controller to automatic m repe~ts per mInute and seconds per repeat,
retaining the original set point. respectIvely. The number in front of the paren-

TUNING SETTINGS AND METHODS

theses in each column is a default setting, a
number that can be used for a download or for a
guess if nothing is known about the loop and no
test can be performed. The numbers within the
parentheses represent a range of typical values. If
'C
your tuning setting falls outside the suggested
--
0
.c :; :; :; :; :; :;
range, this doesn't necessarily mean the tuning G.l .< ..l ..l
U .< .<
..
..l II)
U
~ U U cr. cr. U
U
cr. v
setting is wrong. It does suggest that the loop is c
D
unusual or has some problems such as interac- at

tion, noise, a coated sensor, or a sticky control -

G.l
e=
-;;
=
G.l =, ~= 9 on= ...•~,
N
0 ~
e
VI
e ••••
VI
0

=-, i..
e e e =- e e e ...•e
I
:l VI
Gii 0 VI

valve. The last column shows the tuning meth- IX

!= = =
c
0
VI e ~
Q.

"!
-
0 0
"! VI 0 "!
ods that generally give the best results. The "gen- = = = 'i

eral-purpose" tuning method logic shown in -:; - ~ ~ ~

= on=
or, VI VI
=,
>
.!!
~ Gii
= e= , , ~
e ...•e e ...•e '1- ...•e e
J.
",
e .•.
I
VI
VI J. 'tl
Figure 8 is the closed-loop method (eLM), G.l
~ ~
0 0

-= :: = = = = ...•- = ~
IX G.l ::;)
VI ...•
which is referenced in Tables 2 and 3. While this •
= .Q

= = =
'tl
method is best for gas pressure, reactor, and level oc • ...
loops, it can be used for most other loops as 0.c 9
...•
...•
VI
J.
...•
,;..
~'
0 0
I

9
or.
~
= =. ..
~
...•
'5
o
.r.

well, particularly if the loop's set point is being ~ e e e e e ...•e 0 0

0 0
:::. e e .5
'5
at

on on'"l~ 0
on
VI VI

driven for advanced, batch, cascade, or sequence -:; =:,

0
= = = = ..• 'tl
.r.

control and there are no interaction issues. c c ,

III
•• •• e...• ~= e ~
...•
0
...
'C
-
...• or• - ...•~
or.
~
0 -
.,-.
e e= D
...
~
VI
u
c
at

- or.:::. ...• ...•:::. ...•:::. e

11:.
,::
G.l
~ :::. ~ IS
= - .. ...• ::.. or,
G.l
E
:E

.. ..
G.l
'8
G.l

~.
Co ",G.l",
.. "i:
j
>
:;
",
--..
Co
- -
.. -..
:l
III
G.l

:l
:l
Gii
at
D
I-
--~
~
..
G.l III G.l
~
-
G.l Co Co Co
'C G.l ::;)

..~",", -.. ~.. -- ..

C
..:: :; E Co E

f
0 :l G.l G.l G.l
~ I:T Co l- E I-
Gii ::s
- :; .E - .:
G.l Co G.l
~ 'C 0 III l- e
is. :; .c
", !i G.l 0
CoI
G.l E
:l c
Co I:T llC III III c
< ::s !=
III

~
G.l
IX
G.l
c G.l
is c;
•

TUNING SETTINGS AND METHODS

 .•.
~
III
2.3 The General-purpose Closed-loop
.5 Tuning Method
-e~
~

"'a e
A closed-loop (controller in auto) method has
the following advantages over open-loop (con-
CwS
ta ~
c.El
'tl
Q
.c
'Z .< ::; ~ ~ -
~
{,.l .< ~ .< ::; on troller in manual) methods:

t --e
..J ..J {,.l ..J {,.l
::;
·
{,.l {,.l {,.l v
rI:: rI:: rI::
~ c
"6 l.
~ 1:e -;,
..•-. -. s
..., ...,
S S
::: •...
5'
<=
~
..•~ e
S
5'
<= ~
Q
It forces the user to find the maximum con-
,c
" ~
III '4:l
_<:I'tl c <= ==~
troller gain to minimize peak error and
·
I

.5 .:: e- '4:l
e-..., ! e- <=

... ..• ...

1Il Q
e
I
~
~
_I;/}
1\
::II: ~
fIl
<= <= ...,
<=
...
<= <= <=
- <=
ll.
0
.2 dead time from dead band. It also forces
E~
rJU
-;, 8
<= S S S
8
•...
<=
..•
S
s
<=
<=
Ii
>
.! the user to find out whether reset and rate is
..•-. ~..• ~e
, , even needed, and it ensures rapid set point
..•~-
<=
~I;/}
-'tl
<:I C
'4:l
$ '4:l
<=
'4:l s
~ ~ !
<=
e <=
III
C"'a
.... e
•• fIl Q
<:I U
::II: <:I -- I
~
..•
~
oJ.
~
<=
~
I <=
~
'tl

·
II
::>
response and gets inside the window of
CwS
ta ~
~
'4:l
<=
<=
~
<=
...,
<=
-
<= <=
<=
~ •...
<=
<=
•...
<= <=
'4:l
<=
...,
<= II
.&l
'tl
allowable controller gains for integrating
and runaway loops .
~El oc
..• s~
..• ~.
S S
, •... ;;; ~
'5
~0
III
~e
C e
c. c
Iii ..•~ J.'" oJ.
~ ~ ~
<=
<=
~
g
~
~
~
~
~
, ..•
~
~
S

~
I

=
·
Q
c
2.
·S i" " ""l <= <= ~ ..• ~ '" '"
::>
'2
Loops stay in automatic, which is safer for
~"'a <= wi wi
= = or.
= = ~ difficult or very fast loops.
I;/} ~
Q-S
III
c c
-;
'tl
...,
S ..•, -. or. ;;; S
~ ·
u
c

-u 1Il
u
Q
u ..•~ ~
<=
~
'"
~
or.
~ = ..•~ = ~
"6
Q
3.
=: 1\
rr,; <:I
..•
fIl
-
~ ~
'"
"'!
<=
~
..• ..• ~ ~ ..• ~~ 0
~
or•
~
E
It includes the effects of valve stick-slip and
dead band.
.~~ .. •.... 'S
<:I
£

$0 ••••<:I ~, ~
.. .. ••.. f
::l <:I ::l

.. ••
.. - --..~
<:I -;; ::l
~ <:I
••.. ~ ~ 4. It includes the dynamics and peculiar fea-
~
- ...l ~
"'at! <:I <:I
c.. ~

-
C"'a i- ::>
<:I
e e <:I

- - ~i·
~ :E Q, D"
tures of controller algorithms.
..c..'"'" .e.. .. - -.e..
tll~ c
.:: El .5! Q
li:
::l
:r
::I
~ <:I
.!::!
e <:I
<:I
<:I

•• <:I Q,

-
:3 ii !"'" c
~~
~
Q.
'tl
:; .c :;
<:I e <:I C 5.
••
u
Q, :r llII
'" •
<:I
C u
1Il JE ::I
'0
It includes nonlinearities that are depen-
-
1Il <:I

-
<:I
c
Q~ < :3 !"'" ~ ::II: Z ::II: C
U dent on direction and rate of change.
I
M

-
~
~
~

46 TUNING SETTINGS AND METHODS

TUNING SETTINGS AND METHODS
47
6. It facilitates the tuning of the master (outer)
loop of a cascade system for an oscillating
inner (slave) loop.

This closed-loop procedure uses rapidly decay-

ing oscillations as shown in Figure 9 instead of
the sustained oscillations proposed by the origi-
nal Ziegler-Nichols method. The measured ."
Q

period will be larger for damped oscillations in -=~

industrial applications mostly because of valve ~
C.
dead band, but the error will be in the safe direc- Q
..,c
."-
Q
tion in terms of a larger, more stable reset time , 't:
to
~
setting. The closed-loop procedure should also 0:
.E

-u
III

result in about half of the gain estimated by the

Q •.
-E
0:
1 ~

Ziegler-Nichols method. ~ •.
-E .
0: i .
~
C

.
'ii
III
Q 0:
:: ..
E
c
C

•..:: ..••.
0:
•.
I

Rule 14 - Use the general-pur- -E

•.
-t
~ ..•!
.
0:
c
. . .... pose closed-loop method if the
I
:; !
.z:::
.. =- ... · loop must stay in auto or ifit is
(IS
... .5
•.. ~
c t
5 .E

•..E....
0;"
~ particularly important to maxi- C-' 8
mize the gain for tight control and I ::
•.
QC) c
a fast set point response. However, for dead- ~ ""
time-dominant loops, you should substantially
...•
~
decrease the factor for the reset time. This will ~
prevent the set point response from faltering
because of too much gain action and not
enough reset action. For temperature and pres-
sure loops on exothermic reactors, it is especially
important to use this method to prevent a run-
away.
 A list of steps for the "Closed Loop Method" is
as follows:

1. Put the controller in automatic at normal

set point. If it is important not to make big
changes in the manipulated variable (con-
troller output), narrow the controller out-
put limits to restrict valve movement.

2. For level, gas pressure, and reactor loops

decrease the reset action (i.e., increase the
reset time) by a factor of ten if possible and
trend-record the process variable (PV) and
controller output (CO).

3. Add a PV filter to keep output fluctuations

=cr- c within the dead band of valve caused by
:::'0 nOIse.
- .-
.!
.-
Q
I-

o~
CJ QJ
4. Bump the controller set point and output
and increase the controller gain if necessary
to get a slight oscillation.

5. Stop increasing the gain when the loop

:>~ - 1
•••
starts to oscillate or the gain has reached
your comfort limit. Then note the period.
~~
-=
Q
~
For gain settings greater than 1, the oscilla-
tion will be more recognizable in the con-
•••
QJ
rI:.l troller output (CO). Make sure CO stays on
scale within the valve's good throttle range.
 6. Reduce the gain until the oscillation just
forward, and add rate if there is no inverse
disappears so recovery is smooth. response or interaction.
I

7. For a temperature loop with a smooth ,i

response (no chatter, inverse, square wa:e~ 2.4 The Shortcut Open-loop Method
or interaction), use rate action. If the gam IS The shortcut method is ideally suited for very
larger than 10, reset and rate actio~ are not slow respon~es such as column temperature
needed. If the manipulated flow WIllupset ~here there ISa good control valve and posi-
other loops, decrease the gain or use error- i tlOner and you need to get a quick estimate of
squared control. If a high g~n is used, set a ,
I

!
the controller tuning settings. This method
velocity limit for the set pomt and ~onfig- looks at the change in ramp rate of the OfoPV as
ure the set point so it tracks the PV m man- shown in Figure 10 for about two to three dead
ual and ROUT (DOC). This will enable the times. It doesn't require the loop to be at steady
loop to restart. I
I
state. However, if there is an upset that causes
;
the ramp rate to change, the results will be inac-
8. If rate action is used, set the rate time equal I curate ..!n g.eneral, you should repeat this test in
to one-tenth (1110) of the period and set I both dIrectIOns and use the most conservative
reset time equal to half the period. If rate I
! settings. Also, if the bump in controller output is
action is not used, cut the gain by 50 per- I much larger than the dead band, the shortcut
cent. If the loop is clearly dead-time domi- i
method doesn't include the dead time from
nant, increase the reset action so the reset valve dead band. If the changes in controller
time is about one-eighth (118) of the loop output per scan approach the control valve dead
period. Make another set point change and band in size, you should add the additional dead
adjust the gain to get a smooth response.
Do not become any more aggressive than a
slight oscillation.
I time from the valve dead band to the observed
dead time.

9. If you select gains smaller than 5 for level,

decrease the reset action (i.e., increase the
reset to more than 3,000 seconds), add feed-

52 TUNING SETTINGS AND METHODS I

TUNING SETTINGS AND METHODS
53
 ,
I

!
,
!
I
I

I
,
The shortcut method is also effective for pH
i
i loops because it can keep the test near the oper-
ating point on the titration curve. The use of a
closed-loop method can get confusing for pH
particularly if the oscillations develop into a
limit cycle after being bounced back and forth

--
u
te
tI'.l
between the flat ends of the titration curve. The
period of such a limit cycle is extremely long
and variable, and it will occur for a large range of
~

-=
5
tI'.l
I
>
N

0..

~- c
controller gains.

A list of steps for the "Shortcut Method" are as

.-e .-e
e
z follows:
lo2•• r--

~
-
u

Q"
1. Adjust the measurement filter to keep the
controller output fluctuations caused by
noise within the valve dead band.
~
...= ~
,c
-c
-=
_
Q.
Q = 2. Note the magnitude of output change for
u
bIl .- = r- each reaction to typical upsets. With the
;
.:
~
~a.
c ~
Q

controller in manual near set point, make a

U e:-
'cc
y a. step change in the controller output
I
Q Q.- (~%CO) of about the same magnitude as
••• ~ c
-
t'l Q
Y the output change you noted, but larger
...i ~
than twice the valve dead band .
""" (I)-

~=-
(I)~

=...
•..=-=
l'j .- ~
3. Note the observed dead time and the
change in ramp rates. If the process was
Q..~ lined out before the test, then the starting
ramp rate is zero (~%PV 1 I ~ t = 0).

TUNING SETTINGS AND METHODS TUNING SETTINGS AND METHODS 55

54

,*'
4. Divide the change in ramp rate by the
change in valve position to get the pseudo
I),.%AVP change in actual valve position (0/0)
integrator gain (K;). Then compute the
dead time from the dead band. I),.%CO= change in the controller output (°/0)
DB = dead band from valve hysteresis (Ofo)
5. Use the following equations. For a master Kc = controller gain (dimensionless)
or supervisory loop, omit "tdv- Cd = rate time coefficient (1.0 for baek-
mixed and 0.0 for plug flow volumes)
1(1),. OfoPV 2/ I),.t) - (I),.OfoPV 1/ ~t)\ = reset time coefficient (4.0 for back-
K·
1
= '---------------'\f),.OfoCOI mixed and 0.5 for plug flow volumes)
I<x = gain factor (1.0 for Ziegler-Nichols, 0.5
for IMC, and 0.25 for Lambda)
Kc = --- ~ I<e = open-loop gain (dimensionless)
K·1 • 'td 0 ~ = pseudo integrator open-loop gain
(llsec)
DB I),.%PV = change in process variable (°/0)
'tdv = ~. I),. 0/0 A VP • 'tdo I),. t = change in time (sec)
"t = largest loop time constant (see)
DB "tdo = dead time seen in open-loop test (see)
f),. OfoAVP = If),.OfoCOI- -
- dead time from control valve dead
2 "tdv
band (see)
T·1 = c·I • ('td + 'td 0 ) derivative (rate) time setting (seconds)
V
= integral (reset) time setting
(seconds/ repeat)
 Rule 15- Use the shortcut
method when you want a quick
estimate for a very slow or nonlin-
ear loop, provided that the valve
dead band is 0.25 percent or less
and the step size keeps you near the operating
point. Make sure there are no load upsets during
the test and that you measure the new rate of
change of the PV for at least two dead times.
You should repeat the test for both directions
and use the most conservative tuning.
~
E .5-
.-E-_E

2.5 Simplified Lambda Tuning

If the loop is dead-time dominant, Lambda
tuning is the best method. It also helps to
minimize interactions and suppress oscillations.
Lambda tuning is an open-loop method that is
particularly effective for relatively fast loops,
such as in pipelines, desuperheaters, static
mixers, exchangers, conveyors, spin lines, and
sheet (web) lines, or wherever there is plug flow.
It provides a closed-loop time constant that
approximates the open-loop time constant.
Figure 11 shows the step change in controller
output and the open-loop response of the
process variable. The user simply needs to note
these changes and the time required to reach 98
percent of the final response (T 98). This
simplified form of the equation for Lambda

TUNING SETTINGS AND METHODS

tuning is only suitable for self-regulating 6. The reset setting (repeats/minute) is set
processes. Loop tuning software has expand~d equal to the inverse of one-fourth of the
the actual Lambda tuning rules to cover a wIde response time.
variety of processes.
7. The controller gain is adjusted to be one-
The following is a list of steps for the "Simpli- fourth of the inverse of the open-loop gain
fied Lambda Tuning Method": (Af= 4).

1. Adjust the measurement filter to keep the The equation for reset time setting is as follows:
controller output fluctuations caused by
noise within the valve dead band. Ti = T 98 I 4 = 'tdol 4 + 't (seconds/repeat)
2. Note the magnitude of output change to The equation for gain setting is as follows:
determine the reaction to typical upsets.
With the controller in manual near set
point, make a step change in the controll~r
output (~%CO) of about the same magm- cAr is a tuning factor that is increased to provide
tude as the output change you noted, but a slower and smoother response.)
larger than twice the valve dead band.

3. Note the observed dead time as the time it . ... Rule 16- Use the Lambda tun-
.....
took to reach the first change in the process ..l ing method for dead-time-domi-
variable (~%PV) outside of the noise band. _ . nant or fast loops that can be left
~•. in manual. You should repeat the
4. Note the open-loop gain (the percentage test for both directions and use the
change in measurement divided by the per- most conservative tuning.
centage change in controller output).

5. Note the response time (the time from the

bump to 98 percent of the final value).
 Rule 17- In Lambda tuning, 3.0-Measurements and
decrease A.[to speed up the con-
troller's response to upsets and Valves
set point changes (minimize Ii
PV) and increase Arto slow down
the response and reduc~ l~op i~teracti?n
(minimize Ii %CO). This simplifies tumng to a 3.1 Watch Out for Bad Actors
single knob. The analog/digital converter (NO) chatter from
large temperature measurement spans and short
2.6 Set Point Response and Load scan times is the factor most frequently cited to
explain why derivative action cannot be used in
Rejection Capability
PIO controllers. It is also a considerable source
One way to achieve a good set point response. of dead time because the measurement must get
with a controller tuned for best load response IS out of the noise band if the PIO controller is to
to add a set point lead-lag whose ~agt~me ~raeks discern a true load change from the chatter.
the integral time and whose lead time IS adjusted There is a similar signal-to-noise ratio problem
to be a fraction of the lag time. Another alterna- for level measurements, particularly if the signal
tive is to achieve a compromise by the use of the is noisy due to sloshing or bubbles.
closed loop tuning method with ~bout half. of
the gain that caused quarter amphtu.de cycling The next most frequent problem is sensor coat-
and an integral time (Ti)set as fu.nction of the ing and drift. Resistance temperature detectors
quarter amplitude oscillation penod (fa) and the (RTOs) and smart transmitters can reduce drift
total loop dead time (Ld) as follows: by an order of magnitude. Keeping the velocity
2 that is passing a sensor above 5 fps is the most
T = T /[0.1 +2* {(4* 'td)/(0.7
J a
* To)} ] effective way to keep a sensor clean.

r... Rule 18 - Narrow span ranges

•.
",,; and slow scan times should be
used for temperature and level
~
measurements to minimize the
dead time and chatter from the analogi digital
converter (AID). For level measurements, you
should minimize sensitivity to bubbles,
sloshing, and coating.

~ Rule 19- Use RTDs and smart

~;<~ . Br~~~;:~:;~%~~:1f.~L
through lower maintenance costs
and more accurate process operating points.
";-
--
c~
~c
3.2 Deadly Dead Band VJ

According to pulp and paper plant studies, most

loop variability is caused by poor tuning and
control valve resolution. Figure 12 shows the
dead band that occurs whenever a control valve
changes direction and the staircasing that is the
result of stick-slip action. The dead band and
stick slip are usually largest near the valve seat.
Valve resolution is the minimum change in
signal in the same direction that will result in a
flow change and is usually about half of the
valve dead band. As you approach the resolution
limit there is a dramatic increase in the loop
dead time from dead band, as illustrated in
Figure 13. This additional dead time can be
about five times larger for pneumatic positioners
than for digital positioners. Figure 14 shows that

MEASUREMENTS AND VALVES

it is worse for piston actuators and rotary valves.
Most valve manufacturers will choose a valve •••
position and step size that is near the minimum
~
shown in Figures 13 and 14. For large step sizes, =e
the time required to move large amounts of air .~
into or out of the actuator increases the response
time.
-= e
lrl

Another problem with rotary valves is that the

~
oS ell
WI >
0_
=>
~
5
= .s ell ==
••••
oS .c:
u ~
III
J:~
feedback to the positioner is from the actuator
shaft. What might seem to be minor gaps in key-
tU =.c:
••e :it- -•.=- eIlfe
CIS-_(II)

--
a. - _

lock connections and twisting in long shafts ~= I:

~
~- =-=. =
~ 1:-
actually means that the butterfly disk or ball ~ ~
= ~
may not move, even though the positioner sees
a movement of the actuator shaft. To minimize
~

~
lrl
tU
> ~
~
.c: I:
- =
~
N
WI

Q,

..
III
~ "t'
~ 5
~"t'
~
-
=•.. ...
=
..•
<U
N

e - II V)
this problem, you should use splined connec-
tions and short large-diameter shafts, and you
should conduct actual flow tests on all rotary
~
~
lrl
=
e
a.
'"
- - -=
Q,II
~.c:
WI
-11I_I:
.c: I:
:it ••
~
-~ = ~
5
"t'-"t'
c
CIS
..c
C CIS
•••
Co"t'
<U

=. -
••••
Co
~
V)

valves. All tests on control valves should use step lrl ••• II "t'J:"t'
~ ",-
~ II
=~
sizes that approximate what is expected to occur
.•=:= e,,~
as changes in the controller output from one
scan to another « 0.5%). You should also use
=
~
lrl

==.
-. =.
==
very large step sizes for pressure-relief valves and .~u
for compressor antisurge valves. There is noth- =. "lI:t

=
-
tU
"lI:t "lI:t "lI:t
ing in a valve specification that requires that the ~
valve will actually move. To ensure that a valve Q ~
(II)
(j
~

-..
I C (II)
will respond to meet the needs of a control loop, f'l')
=-
Co~
you should add the dynamic classes in Table 4
to the valve specification sheet when purchasing
~
~
~ .-
~E-
5
(II)

a control valve.
""
 Table 4 - Dynamic Classes of Control Valves (Four
classes A ~ D for each of the four categories 1 ~ 4)
I-Minimum Step 2-Maximum Step
Classes Classes
Class A - 3.0% ± 0.3% Class A - 5% ± 0.5%
Class B - 1.0% ± 0.1% Class B - 10% ± 1.0%
Class C - 0.5% ± 0.1% Class C - 20% ± 2.0%
Class D - 0.2% ± 0.1% Class D - 50% ± 2.0%

3-Response Time 4-Minimum Position

Classes Classes
Class A-IS sec Class A - 30%
Class B - 5 sec Class B - 20%
Class C - 2 sec Class C - 10%
Class D - 1 sec Class D - 0%

Note: The response time is the time it takes the trim

(not the actuator) to stay within 10 percent of step
or within 0.1 percent of span, whichever is largest.
(Overshoot is OK if recovery is within this offset
from the desired position of the trim)

Referring to Table 4, consider a DACB class

control valve. It will respond to signals larger
than 0.2 percent and smaller than 5.0 percent in
less than 2 seconds above a position of 20 per-
cent. If you don't care about how well a valve
responds, specify class AAAA-almost any valve
will meet it.

A general-purpose application would be BAAB.

Examples of other classes for various types of
control loops are given in the following list. To then attach a travel indicator to the butterfly
choose a typical class, use the loop type that disk or ball to check whether they track the
actually throttles the control valve. For cascade changes in the actuator shaft.
loops, you should use the slave loop.
Finally, the change in the flow measurement for
• Vessel or Column Temperature Control a change in controller output will also show if
>CABC the control valve is on a portion of the installed
• . Exchanger Temperature Control > characteristic that is too steep or flat. If there is
CACC no flow measurement, ask the valve manufac-
Pipeline Temperature Control > CCCC turer to generate an installed valve characteristic
Liquid Pressure or Flow Control > CACC for your piping system. Signal characterization
Vessel Level Control > CABC to linearize the valve will decrease valve dead
Gas or Steam Pressure Control > CBDC time from its resolution limit for the flat portion
Compressor Surge Control > CCCD of the curve, but it will increase this dead time
Vessel pH Control > DABD for the steep portion of the curve.
Pipeline pH Control > DBCD
Pressure Relief > BCCD Figure 15 summarizes the maintenance tests for
For split range, use class D for last category a good or bad valve. The change in PV can be
(minimum throttle position) any process variable, but it works best if it is a
fast measurement such as flow or pressure. Slow
The best way to see if the flow actually changed process variables such as temperature and level
for a small change in controller output is to respond much too slowly, especially for small
make a sensitive flow measurement with very changes in controller output, to make it practical
low noise. This is particularly important for to determine valve dead band.
rotary valves or pneumatic positioners. Another
clue that the valve has a problem is if the control Rule 20 - To achieve the tightest
loop oscillations are smaller in manual than in control (i.e., least variability), use
auto (this assumes you have tuned the con~rol- sliding stem valves and digital
ler). If it is a critical rotary valve and there IS no positioners. If you use rotary
flow measurement, you need to take the valve valves, you should have a flow
out of the pipeline. In the shop, you need to
 measurement, splined shaft connections, and a
short large-diameter shaft.

."
g~
•• >
..
i;J g£±~{~!~~lf:i~::
more sensitive. Be careful to avoid
increasing the valve dead time from dead band
on the steep portion of the curve. Also make
sure you don't eliminate an equal-percentage
characteristic used to compensate for a flow ratio
!!
"'i=-
gcun.
! ~= j
u c:"•• -l-
_ .... ..
=• .•
-
.z:
In the normal scheme of things, slip is worse
:: " than stick, stick is worse than dead band, and
dead band is worse than stroking time. For slid-
ing stem valves, stick-slip goes hand in hand
with dead band since the common cause is
excessive packing friction. Rotary valves can
have a large dead band from gaps in connections
and linkages and not much stick-slip. However,
high performance valves generally have signifi-
cant stick-slip and shaft windup from excessive
packing and seating friction. Stick-slip causes a
limit cycle in all loops, dead band causes a limit
cycle in integrating loops, and a large stroking
time causes instability when the rate of change
of the controller output from integral action is
faster than the slewing rate of the valve.
 3.3 Sticky Situations In rotary valves, the positioner feedback mea-
For those of you who have been stuck with dead sure~ent is.often t~e actuator shaft. The tight-
band and stick-slip and want to break free, this ness of the Intervenmg connections determines
section will discuss the sources of the problems whether the po~itio~er is actually seeing what
and details of the solutions. the ball or dISCISdomg. On top of this, there
may be a scotch yoke actuator where the loose-
. Tight connections are the goal for the transmis- ness ?f the gear mesh adds another degree of
sion of force from the actuator to the internal slop m valve response.
ball, disk, or plug. For a sliding stem valve there
is a single connection between the actuator shaft .. <., Rule 22 - To reduce the extra
and the stem of the plug. In a rotary valve there i/'CSJ de~d hand in rotary valves, use
is a connection between the actuator shaft, valve sphned shafts and digital position-
shaft, and ball or disk stem. There may also be ~ ers that measure the actual rota-
linkages associated with the translation of verti- tion of the internal hall or disk.
cal actuator motion to rotary valve motion. The
play in each connection, which is usually a For surfaces that must slide, the tightness and
pinned, keyed, or splined slot, determines how r?ughness of these surfaces leads to high fric-
much dead band (backlash) exists whenever the tIOn, commonly called sticktion. The smallest
valve needs to reverse direction. While the play change in signal that a control valve can respond
of loose slots may be fun at Vegas, they are to regar~less of its direction or past movement is
deadly in control loops. called suck: When the valve does move, it slips
where the mcr~mental change is approximately
Manufacturers have been doing a better job of ~qual to the st~ck.Excessive break away torque,
not only improving the tightness of connections madequately SIzed actuators, operation near the
but also in increasing the number of pins, keys, s~at, and poorly tuned positioners can result in a
and splines around the circumference of the slIp J?uch .la~ger.than the stick even though the
shaft to increase the effectiveness of the trans- p~eking.fnctIOn ISnot too high. Control loops
mission of force. A splined shaft with a large WillOSCIllateregard!ess of tuning with an ampli-
number of tight spline connections has the least tude eq~al to the slIp ~ultiplied by the open
backlash. loop gam. The resolutIon capability or sensitiv-

MEASUREMENTS AND VALVES

ity of the valve is set by the slip, eve~ though it 4. Plunges in rotary torque requirement upon
is commonly defined based on the sock. break away when the disc jumps open
Here is the rub: Until valves can be made with- 5. Tightened or roughened packing surfaces
out sliding surfaces, there will be friction and a
stick and slip that are at least as large as the man- 6. Low performance or poorly tuned position-
ufacturer's stated resolution or sensitivity limit. ers
For the best designed valves in perfect condi-
tion, the resolution is about 0.05%, which An isolation, safety interlock, or batch sequenc-
approaches the resolution li~it of a.12-b~tana- ing valve designed for tight shutoff should never
log/ digital (AID) converter wIth a sIgn bIt. How- be used for a control valve and vice versa. On-
ever, most valves are far from perfect and the off valves do not make good throttling valves
environment in process applications is not any even though a positioner has been added. On-
where as nice as in the valve shop or test lab. off valves are designed to be either completely
Consequently, users have lowered their expecta- open or closed. Control valves are designed to
tions and consider 0.5% resolution to be a good move freely near the closed position. Control
valve. The following situations can make reality valve specification sheets that have a leakage
reek by increasing stick-slip and achieving even a class but not a stick-slip requirement, leads to
0.5% resolution seem like nirvana. the selection of on-off valves.

1. Crud on the stem, trim, or seals that reduce While it is best to reduce the sources of high
the slickness of surfaces friction, the stick-slip can be reduced by install-
ing an actuator with plenty of muscle and a digi-
2. High temperatures where the expansion tal positioner tuned with a high gain and rate
reduces the clearances between surfaces setting.

Shaft windup in rotary valves where the

Rule 23 - To reduce stick-slip,
shaft twists and then springs free Gumps) .
. ..;};
".""~"
.
.."'." avoid valves designed for tight
shutoff and use low friction prop-
~ erly tightened packing, gener-
 noise (best checked in manual) and of oscilla-
ously sized actuators, and aggressively tuned
tions for upsets will be smaller for a fouled sen-
digital positioners. sor. The appearance of a flatter trend has fooled
many individuals. A biochemist once showed
3.4 Fouled Sensors me how he smoothed out the trend chart of a
A coating on a thermowell will gene~allyreduce fermentor temperature by partial withdrawal of a
its heat transfer coefficient and provIde a ther- temperature sensor in its thermowell. The result-
mal insulating barrier to the fluid temperature. ing air gap acted as an excellent thermal barrier.
The response of the sensor slows do~n.and .the Spring loaded temperature sensors are used to
temperature difference betwee~ the mSIde op ensure the tip of the RTD or thermocouple
and process fluid increases,.whICh ~ranslates to a touches the bottom of the thermowell.
bigger lag time and dynamIC err.or m the temper-
ature measurement. For a rampmg temperature .... Rule 24 - If the oscillations in
change, the measure? temperat.ure will b~ . a loop become more frequent
delayed by this lag orne. Also, If ~e la~ tIme IS _ and slower for the same tuning,
more than 25% of the total dead orne m the ~ cheek to make sure the sensor
loop, a controller tuned for performance rather response has not slowed down
than robustness may go unstable. The ~ontroller from a coating.
gain must be decreased and th.e.reset orne
increased to deal with the addItIOnal sensor lag A pH electrode is incredibly sensitive to coat-
ings. A 10 millimeter thick film can increase the
time.
lag time from 5 seconds to 100 seconds. The
Since temperature control syste.mst.end to have coating acts as a barrier to the migration of
too small of a reset time to begm WIthper the hydrogen ions that need to get the surface of the
Figure 2 heating and cooling example, ~e most glass electrode to change its millivolt potential.
noticeable symptom for a fouled sensor ISa reset The coating can be so thick that the electrode
cycle and a longer period. However, t.he sensor does not respond at all.
lag acts like a signal filter or a tr.ansmltte~ damp-
ening setting and attenuates nOIse and d~stur- The time to about 63% of the final response for
bances seen by the loop. Thus, the amplItude of the insertion of a temperature sensor in a bath or

MEASUREMENTS AND VALVES

an electrode in a buffer is the sensor lag time. cult to remove and increase the susceptibility to
Washing or decontamination of the sensor and further coating. The key is to never let the coat-
bath or solution itself may remove most of the ing start. The installation of thermowells and
coating. Thus, removing the sensor for testing in electrodes in recirculation lines is the best insur-
a shop often does not disclose the severity or ance because these lines tend to have a constant
nature of the coating. If safety procedures allow high flow rate. If the fluid is abrasive, the veloc-
it , the sensor surface should be inspected as it is ity may need to be lowered by the addition of a
removed. wide spot in the line.

Rule 25 - If safe and practical, ~ Rule 26 - Install thermowells

visually inspect sensor surfaces
for coatings as they are removed
and before they are washed and
t.,..~
.
~~:~~:l::~~~:r
:::ne;~:~~;~
the process flow stops, dram and
sent to the maintenance shop. flush the line to avoid having the
sensor sit in a stagnant process fluid. Recircula-
Coatings are less likely to occur if a high fluid tion lines are more likely to keep the velocity
velocity sweeps parallel to the surface. The . high and constant at the sensor.
threshold velocity that is reported to make a dls-
cernable difference in the rate of coating is Flat surface electrodes get the same cleaning
about 5 feet per second, which is typical for a action for a much lower velocity. The primary
pipeline at normal flow. The ~elocity in the flow pattern should be parallel to the surface for
most highly agitated reactors IS rarely more than the best sweeping action.
1 foot per second. Higher velocities also reduce
the lag time of clean thermowells and electrodes. 3.5 Noisy Measurements
Thus thermowells and electrodes installed in
pipelines rather than vessels stay cleaner and If the sensor location is too close to the dis-
have a faster response clean or dirty. However, if charge of an exchanger, desuperheater, pump, or
the flow is throttled or stopped, coatings will static mixer, the response may be noisy because
develop that may not be able to b~ cleaned .by a streams have not had the time to mix or bubbles
return to normal flow. Sticky coatmgs are diffi- or solids time to dissolve. On the other hand,
the distance between the equipment and the
Rule 27 - Install thermowells
sensor adds a transportation delay equal to this. and sensors within 5 to 10 pipe
volume divided by the flow. Therefore, the opu- diameters of the equipment dis-
mum location is about 5 to 10 pipe diameters
charge with the tip near the center
downstream of the equipment. Strong acids and of the pipeline to minimize noise.
bases require twice the distance because of the For a thermowell, the best location is an elbow.
extreme process sensitivity (gain). ~ubbles and A pH electrode must be angled above horizon-
tal. Bubbles and strong acids and strong bases
particles may require even .long~r.dlstan~es. The
bubble size decreases and Its abIltty to dIssolve require longer distances.
increases as the velocity increases.
Differential head, thermal mass, and vortex flow-
Sometimes overlooked is making sure the tip of meters will have noisy signals unless the flow
the sensor is near the middle of the pipeline to profile is uniform. The number of pipe diame-
see a more representative and smoother te~pera- ters of straight run needed upstream can vary
ture or concentration. This is particularly Impor- from. 10 to 40 pipe diameters depending on the
tant for pipes with bubbles, particles, high . locatIOn of valves, fittings, and elbows upstream.
viscosity, or jackets. Installing a thermowell III Flow straightening devices can significantly
an elbow keeps the insertion length in the center reduce the straight run requirements. A control
of the pipe. Since electrodes don~t ~a~e conduc- valve should never be located upstream of a
tion errors like thermowells, maxImlzmg the flowmeter. The straight run needed downstream
insertion length in the process is not important. is ge?eral~y 5 pipe diameters. The straight run
However, pH electrodes require that an elec- speCIficatIOnsfor magnetic flowmeters are just a
trode be installed above horizontal so that the couple of diameters. Corriolis flowmeters have
internal air bubble does not get lodged in the ~o official requirement, but it is not a good prac-
tip. Most manufacturers designate a.30 to 45 tIce to bolt them to control valves or equipment.
degree angle from horizontal as opumal. .
... '.- Rule 28 - Ensure differential
". .~... head, thermal mass, and vortex
... fl~wmeters have enough straight
~ pIpe runs upstream to see a good

MEASUREMENTS AND VALVES

1
flow profile that reduces noise. Always install 4.0-Control Considerations
the control valve downstream of the flowmeter.

Operating near the low limit of the, allowable

range for the sensor can cause a no~sy resp,onse. 4.1 Auto Tuners
This is particularly a problem for dIfferentIal The relay method auto tuner shown in Figure 16
head flowmeters. Changes in Reynolds numbe~ is effective and simple. It keeps the loop under
can change the meter coefficient, and. the tranSI- control by constantly forcing the PV back and
tion zone between turbulent and lammar flow forth across the starting point of the test. Its
can cause erratic behavior. If you go below the effectiveness depends upon the noise band
limit for vortex or magnetic flowmeter, the m~a- being set correctly so that fluctuations from
surement signals can go bonkers. The rangeabil- noise are not measured as loop oscillations. fig-
ity of vortex meters and magmeters dep.ends o~ ure 17 shows a single step in controller output
the fluid velocity in the meter and the kinematIC that uses the shortcut tuning method to provide
density and conductivity, respectively. an estimate of tuning settings and then a series
of successive steps to provide an automatic iden-
Rule 29 - Make sure the meter tification of valve dead band.
_ _---. size and range for the fl~id condi-
, tions ensures the operanng range
4.2 Uncommonly Good Practices for
~- is always above the low rangeabil-
ity limit of a differential head, Common Loops
magnetic, thermal mass, and vortex flowmeter.
For flow and pressure loops, first make sure the
scan time is fast enough. For gas pressure, the
gain must be maximized. For level loops it is
most important that you decide whether the
level control must be tight or loose and that you
minimize the reset action. For temperature loops
it is critical that you use a narrow calibration
span and slow scan time. For concentration
loops you must make sure that the sensor is not
 Figure 16 - Relay Method Auto Tuner

Ultimate Gain Ultimate Period

Measurement
(%) Ku=--
4·d
I. T
u
.\

n·e

Output (CO)
(%)

Time
(Olin)
e = sq rt (a2 - n2) If n = 0, then e = a
alternative to n Is a flIter to smooth PV

Measurement
(%) Ramp "me Identification
and shortcut estimate of
controller tuning seHings

Output (CO)
(%)

Time
(sees)
coated (velocity> 5 fps) and that the signal is have Scotty and warp drive, you are stuck with
fast (minimum transportation delay) and the dead time caused by equipment, piping,
smooth (no noise). For pH loops with an operat- instrumentation, and control valves in your
ing point on a steep titration curve, the greatest loop.
need is for an exceptionally precise control valve
(resolution < 0.15%). Characterizing the signal A dead-time compensator is very sensitive to an
of the process variable according to the titration overestimate of the dead time. A dead time that
curve can help reduce the oscillations. is 25 percent larger than actual can cause insta-
bility. A major source of dead time is a transpor-
4.3 Dead-Time Compensation and tation delay, which can be computed for a
pipeline as the volume divided by the flow or for
Warp Drive a sheet as the distance divided by the speed.
When I first left home, my dad said, "Be as hon- Overestimates of the process gain and underesti-
est as the day is long, don't talk when you mates of the time constant are also problems but
should listen, and don't be fooled into thinking to a lesser degree.
a dead-time compensator can eliminate process
dead time." A dead-time compensator such as a To derive the full advantage of a Smith Predic-
Smith Predictor can cancel out the effect of dead tor, you should increase the controller gain and
time for changes in the controller output and the reset action. For a negligible time constant,
make possible higher gains and faster reset the reset time can be set equal to one-fourth of
action (smaller reset times). However, it is a the uncorrected dead time to provide a reset
common misconception that it eliminates dead action that is an order of magnitude greater. A
time when correcting disturbances from feed- dead-time compensator is most effective on
back action. The minimum peak error still corre- dead-time-dominant loops where the dead time
sponds to the excursion of the process variable can be accurately calculated and updated in the
in one total loop dead time. The disturbance predictor.
and correction must still make a complete tra-
versal of the block diagram. Dead time in the Most dead-time compensators can be reduced to
plant cannot be eliminated by an algorithm some form of the Smith Predictor shown in Fig-
without violating the laws of physics. Unless you ure 18. The PID controller output passes
through a dead time, single time constant, and
steady-state gain model of the valve, process,
and sensor dynamics. The model output with
and without the loop dead time is subtracted
from and added to the measurement of the pro-
cess variable, respectively. This leaves the model
output without the dead time as the controlled
variable. The dead time has been removed from
the loop as far as changes in the controller out-
put are concerned. Note that the controlled vari-
able is no longer the actual process variable.

....w Rule 30 - To maximize the sta-

.... -i, bility of a dead-time compensator,

~
.
update the dead time as quickly
and accurately as possible and -~
o
u
is
GI
make sure you always underesti- ~
a..
mate the dead time and process gain and over-
estimate the time constant. You should include -E
~

V)
the change in dead time resulting from a change
in a transport delay by using a computation in
which the dead time is inversely proportional to .
.............................. ,.
~
flow or speed. -'0 ~_
GI 'tI

J:~
c:: u
o~
u
 Rule 31 - The reset time should
be set to be about one-fourth (1/4)
of the uncompensated dead time
to get the most out of a dead-time
compensator. The reset time can
be decreased by an order of magnitude by add-
ing a dead-time compensator with a 10 percent
or less underestimate of the dead time.
I> I>

4.4 I Have So Much Feedforward, I Eat

~I> E
~ E
~ •E
b() i ~
Q.

before I Am Hungry =
.~
A feedforward signal that is accurate both in F
gain and in timing can make an impressive
-=•••
t'=
impact on control loop performance, especially ~
for unit operations with large dead times such as :a
~
...> ...> ...
>

distillation columns and large reactors. Ideally, ~

-..
the compensating effect from the feedforward I
CI"
signal should arrive in the process at the same
~
time as the load change and be equal but oppo-
site to the upset. For fast loops, the timing is ~
tight, and the feedforward signal may arrive early •• I>
E
~
I>
E •E
~ ~
and cause an inverse response. If it is too late,
the feedforward signal creates a second distur-
bance. For fast loops, the timing is more critical
than the gain, but dynamic compensation is
often neglected. Figure 19 shows the patterns of ...> ...> ...>
response that are a clue to how you should
adjust the feedforward gain and timing.
You can greatly improve a loop's set point 4.5 Cascade Control Tuning
response by adding the percentage change in set Cascade control is a type of control in which a
point into the controller output as a feedforward seconda.ry (slave or inner) loop is added that gets
signal that has a gain of about half of the con- a set pomt from a primary (master or outer)
troller gain, an action opposite to the control l?op. If th~ secondary loop response (both dead
action, and a slight filter. This provides the kick tIme a.nd bme constant) is five times faster than
you would normally get from a high gain setting. t~e pnmary loop response, there is no interac-
The loop can then rely on reset action to make tIon between the loops, and the secondary loop
the rest of the transition to the new set point. can correct for upsets it can measure before they

i3.~.~:,~
..
!:ili~d~n~~i~e~~:;;:~.
tion (dead time and lead lag) must
. be set properly to get the most
affect the primary loop. If ~he secondary loop is
not fast enough, you must mcrease the scan time
a~d the PV filter time of the primary loop or you
I!
WI have to dec:ease the controller gain and rate
a.ctlOno~ the pnmary loop to reduce the interac-
benefit. For fast processes the tim- tlOn. BeSIdescatching an upset quicker, the sec-
ing is more critical. It is better for the signal to ondary l<:>opmay also help linearize the response
arrive a little late than too early because inverse of the pnmary loop. For example, in the cascade

.
response is extremely disruptive.

Rule 33 - Add a feedforward sig-

of reactor temperature to coolant jacket exit tem-
peratu~e, the secondary loop of jacket tempera-
ture WIllmake the process gain for reactor

ij
... nal of a set point in percent to the
.. L,
temperature linear in addition to sensing coolant
controller output with a gain temperature disturbances before they affect the
about half of the controller gain, re~ctor temperature. Similarly, the cascade of
an action opposite of the control- exit coolant te~perature to a flow loop on cool-
ler action, and a small filter. The feedforward ~nt makeup WIllremove the nonlinearity of the
action is opposite the controller action because mstalled characteristic and dead band of the
the controller action works on error, which is the valve in addition to reducing the effect of cool-
measurement minus the set point. For nonlinear ant pressure cha?ges. The most common type of
valves, you may need to add signal characteriza- seco~dary loop ISthe flow loop. It can consider-
tIon. ably Improve the performance of a primary loop

CONTROL CONSIDERATIONS
for concentration, level, and temperature con-
The most common mistake is to forget to prop'
trol. The flow loop is not generally recom-
erly set the output limits of the primary loop.
mended for liquid pressure control because the
You must set the output limits on the primary
speed of response of the liquid pressure and flow
loop so they match the set point limits on the
are about the same. Another type of cascade
se.condary loop. Also, the primary loop must not
control that should be used more frequently is
WInd up when the secondary loop output is at its
the cascade control of still, reactor, or evapora- output limits.
tor temperature to stearn pressure control. In
this kind of control, the stearn pressure loop
~ Rule 34 - If the dead time and

i~;~!~i:~~;~~5~:
compensates not only for steam supply pressure
upsets but also for changes in the condensing
rates (heat load and transfer) as reflected in the »- >~

stearn coil or jacket pressure.

the primary loop must be slowed
down. The scan time and filter time must be
You should tune the secondary loop for an
increased or the controller gain and rate action
immediate response before tuning the 'primary
must be decreased in the primary loop.
loop. An offset in the secondary loop ISusually
of no consequence since the sole purpose of th.e
secondary loop is to meet the demands of a pn-
.. .... Rule 35 - You must tune the sec-
. ·v> ondary loop first and use gain
mary loop. You should tune the se.c~ndary loop
.-:. and/or set point feedforward
with maximum gain action and mInImum. reset
~ action in it so the secondary loop
action and use a feedforward of the set POInt.
This is particularly important for cascade control can immediately respond to the
set.poi~t changes from the primary loop. Reset
of gas pressure control to flow. If the abrupt
actIon IStoo slow to be the major source of con-
changes in the secondary loop output upset
trol action in the secondary loop. If you use a
another important loop, then you may need to
feedforward of the set point, make sure it is
use Lambda tuning to make the secondary loop
scaledyroperly. If the secondary loop fights with
response smoother and more gradual.
ot?~r I~porta~t loops, use Lambda tuning to
mInImIze the Interaction.

CONTROL CONSIDERATIONS
 Rule 36 - Make sure the output 5.0-Troubleshooting
limits of the primary loop match
the set point limits of the second-
ary loop. In a fieldbus-based sys-
tem, the primary loop output limits 5.1 Patience, Heck, I Need to Solve the
use the engineering units of the secondary loop. Problem
In older systems, the primary loop output limits
Figure~20 and 21 show some diagnostics for
are usually expressed as a percentage of the sec-
loops m.manual and auto. In these diagrams,
ondary loop scale. reset actIOn (repeats per minute) is referenced
rather than reset time (seconds per repeat). If
4.6 Keep the Secondary Loop on the there are fewer oscillations with the loop in man-
Move ual, the problem is either a poor control valve or
controller tuning. If there are also fewer oscilla-
A common mistake is to tune a secondary loop
tions in other loops, then they are probably
for a smooth set point response in a cascade
caused by an interaction between this loop and
loop with a large process time constant. The .
other loops. If the oscillations persist and are
load rejection capability of the secondary loop IS
~ast,they are probably due to electromagnetic
much more important. In particular, it is the rise
mterference (EMI), sensor noise, pressure waves,
time rather than the settling time of the second-
or resonance. If the oscillations persist and are
ary loop that should be minimized to make the
slow, then they are periodic upsets from other
initial response as fast as possible. The remote
loops that have poor tuning or valves or are
set point (RSP) of a secondary controller is often
caused by on-off actions (level switches), steam
changed by the primary controller before the
~raps, pressure regulators, burps (column flood-
secondary loop PV even crosses its RSP. The use
mg), or flashing~ If the oscillations dissipate
of a filter on the RSP of a secondary loop will
when the valve IS closed, the oscillations were
seriously degrade cascade control. caused by pressure fluctuations at the valve.

With a loop in manual, you can find the valve

dead band by making a step change of 0.25 per-
cent in the valve and then waiting more than the sure loop, on-off actions, or oscillating users on
dead time to see if the actual process variable
responds to the change in the controller output.
For slow processes or in situations where you are
reasonably sure the valve is really lousy, you can
increase the step size to 0.5 percent or even 1.0
13
the same header.

1
1.'•'.,'..:.'.;.\
'.. .
RJtle 39 - To track down the
\:;,t.(ii. i •.. source of an oscillation, put each
loop in manual and stop each on-
percent. This will allow the test to be completed ',.•....~ ...> off action one at a time. When the
before your patience wears thin. If there is no oscillation stops, you know the cul-
response, repeat the step. The steps should all be prit was the last loop you put in manual or the
in the same direction until there is response of last on-off action you stopped.
the process variable that is outside of the noise
band. Then repeat test for the opposite direc- If the loop is in automatic and there are fast peri-
tion. The absolute magnitude of the total num- odic upsets, the oscillations are probably caused
ber of step changes needed to get a response in by EMI, sensor noise, pressure waves, or reso-
both directions is a measure of the valve dead nance. If you experience slow periodic upsets
band. and a period much greater than four times the
dead time, then you should suspect on-off
Rule 37 - If the oscillation goes actions, steam traps, pressure regulators, burp-
away when you put the loop in
.............•
·....L:. ,.,...,.. ing, or flashing as the source. Otherwise, tuning
.v_ manual, then the loop is the cause is probably the culprit. When PV wanders in
~ of the oscillation. The culprit automatic it is a symptom of the control loop
could be the control valve, tuning, operating on the flat portion of an installed
or loop interaction. valve characteristic. If the PV falters for a dead-
time-dominant loop (see Figure 3b), it is an indi-
_ Rule 38 - If the oscillation only cation of too much gain action. If the PV stair-
, . . >;.L. goes away when you close the cases (see Figure 5), it is an indication of too
valve, then it was caused by pres- much rate action. If the PV overshoots and
~. . sure fluctuations at the valve. The develops a slow oscillation whose period varies
culprit could be an oscillating pres- with the reset setting (see Figure 4), it is a sign of
too much reset action (i.e., the reset time too
small). If the loop period is very sensitive to reset
action, it is a sign that there is excessive valve
dead band. For sticky valves, you can make the
co
oscillation significantly faster by using more rate c
".°i'"y ~
'" :C.f!:.c "c
" :E.,S!:c
and gain action. This will get the controller out- u
;>
u
;>
" "~
.9 .9
!-
co ~ s" ~
c -u
._ ] c ti u
put through the dead band more quickly. a;
;>
a;
;> Ol '"
tlO~
y

e E
Ol
.~ ~ E
.J

If the period of an upset is near the natural u

E .Eu g
period of the loop, which is two to four times 1
the dead time, then the loop will amplify the
disturbance, and the oscillation will be worse ~
when the loop is in automatic. If the period is .5
lfl
Cl.
much less than the natural period it is uncon- ~
trollable noise. For a loop to have a chance of j
attenuating the upset, the period of the upset ~
•• C" c
u
00 "ij
v-
co
t: .r:; ~ .E :c .E
must be more than twice as large as the natural lfl
U
~ ... u

'': '"
"C -=:; ~ ~ .9c
period of the loop. lfl
~ ~~
u
/;;. "
c
";;; ~ u
co ~ E
"
v- ~
u
c Ol
;-
~ :;; :a
Note that on Figures 20 and 21 "reset too hi"
means the reset action is too high, which means
is
I
the reset time is too small. Q
N
~
••
.. Rule 40 - If the oscillation
..•... . ••••.• period for the loop in automatic
..
~
~

drastically increases for less

~ aggressive controller tuning, the
valve has excessive stick-slip. It
takes longer for the controller output to work
through the resolution if the PID action is
slower due to less gain, reset, and rate action.

.~
 The oscillation can be made much faster and
tighter by using higher gain and rate action.

Figure 22 is a spectrum analysis for a level,

which shows that the predominant frequency is
0.08 cycles per minute. By analyzing the power
spectrum of process variables in the application
depicted in the figure, Walsh Automation was
able to find the source of a cycle with a twelve-
minute period in a series of distillation columns.
It was an oscillation in the feed, which was
caused by level switches on a feed tank that
turned pumps on and off. This power spectrum
analysis tool can quickly point to common peri-
--c:;
."~ .=:c~
odic oscillations that noise and upsets would
-5 §~~ make it difficult to spot in trend recordings.
g fJ
.~~ ~
tlCI ~ •.•

~
;- Rule 41 - A power spectrum ana-
.:. lyzer can rapidly point you to the
culprit by indicating which loops
~- experience significant peaks in
the power at the same frequency.
The first step is to enter the data for loops in
automatic into the power spectrum analyzer.
The data gathering must be done quickly
enough to prevent aliasing. For chemical pro-
cesses,it is sufficient to use data from a historian
with no compression and an update time of 1
second. For sheets (webs), you may need to store
the data using a device with an exceptionally fast
 scan time (50 milliseconds or less) that is directly
connected to the controlled variable terminals.

5.2 Great Expectations and Practical

Limitations
When all is said and done, a loop will at best
only pass on variability from its PV to its output.
Variability in the output of one loop can greatly
disturb another loop. Loops will actually
increase the variability of the PV if there is a
sticky valve, overly aggressive tuning settings, or
upsets that have a period close to the natural
period of the loop (two to four times the loop
dead time). Most periodic upsets are caused by
on-off actions and oscillating loops (which are
often associated with level switches or level con-
trollers that have too much gain or reset action).
You can avoid a lot of headaches and wasted
effort by first resolving valve problems and peri-
odic upsets.

The most troublesome practice is using reset

action correctly. It is not widely understood that
the reset time as a factor of the loop period var-
ies by an order of magnitude depending upon
the degree of dead-time dominance and self-reg-
ulation. For example, for a process that tends to
ramp or run away, the reset time should be larger
than ten times the loop period. For a self-regulat-
ing process with a small dead-time-to-time-con-
exothermic reactor temperature loops to protect
stant ratio, the reset time should be between 0.5
against instability caused by reset. In practice,
and 1.0 times the loop period. For pure dead-
some reset is useful to eliminate offset and facili-
time processes, the reset time can be one q':lart:r
tate startup. For level loops, it has been found
(1/4) of the dead time. Since the ~atural p~nod IS
that to prevent oscillations, the product of the
twice the dead time for a dead-ome-dommant
controller gain and reset time must be greater
loop, the reset time ends up being about o.ne-
than 4 times the fastest time for the level to
eighth (1/8) of the loop pe~iod. If a dead-o~e ramp full scale. While the actual minimum
compensator is added to thIs loop, the reset orne
product of gain and reset that triggers oscilla-
can be set to be about one quarter of the uncom-
tions will change depending upon the amount
pensated dead time. If the dead ~me is underes-
of loop dead time and valve dead band, the rela-
timated by 10 percent, the reset orne can ~e as
tionship is useful to determine how to simulta-
low as one eightieth (1180) of the loop penod
neously adjust the gain and reset for level loops.
for a dead-time compensator.
If the loop response is smooth and the control-
ler gain is doubled, the reset time can be halved
To make it even more interesting, the rule of
and still keep the product of the two adjust-
thumb to set the reset time greater than ten
ments the same. It is critical to note that the
times the loop period for integratin~ and ~n~
opposite is true (the reset time must be doubled)
away processes is designed to essentially elImI-
if the loop is approaching the upper gain limit. If
nate the possibility of oscillations due to reset
you are confused, don't feel alone. It is safe to
action. There is a gain window for such non self-
say that 99% of users who don't carry this guide
regulating processes. If the gain is b~low the low
don't even have a clue that this strange twist of
limit or above the high limit, there ISa loss of
the rule on tuning reset even exists.
control. The closer the controller is to the low
gain limit, the more sensitive it is to reset action.
The following equation by Harold Wade shows
Proportional action makes the response more
how to eliminate oscillations near the low gain
self-regulating and thus, more tolerant of reset
limit for a level controller if the dead time and
action. Some consultants recommend propor-
valve dead band are negligible:
tional-only controllers for level loops and pro-
portional plus derivative controllers for
 Rule 42 - The update time of
trend recordings should be less
Where: than one tenth of the loop dead
time or time constant. The com-
LlF = maximum difference between inlet
pression of data should be less than
and outlet flows (gpm) one tenth of the step size and less than the
Kc = controller gain measurement or valve resolution, whichever is
T1 = controller reset time (minutes/repeat) largest.
TL = fastest level full scale ramp time
(minutes)
Vs = volume for level measurement span
(gals)

Finally, in order to tune the loop you need to be

able to see the response. If the update time is too
slow, the period of an oscillation and any time
measurements will appear longer than actual. If
10% accuracy is desired, then the update time
must be less than one tenth of the dead time or
time constant. The scan times of any devices
that supply data should be twice as fast as the
trend update time to ensure sufficient over sam-
pling. In order to get good accurate estimates of
gains, the compression (value for exception
reporting) should be smaller than the measure-
ment or valve resolution. Otherwise the gains
will be in error to the degree that the compres-
sion approaches the step size.
6.0- Tuning Requirements
for Various Applications
This section hits the highlights of what are the
major considerations in tuning loops in actual
applications. The biggest most common prob-
lem is that the more critical control loops in the
process industry are slow (long dead times and
large time constants). The tuning tests for these
loops take so long they are vulnerable to unmea-
sured load upsets during the test. Particularly
problematic are large distillation columns where
the time to steady state exceeds a shift and batch
operations where there is no steady state or the
time to steady state approaches or exceeds the
batch cycle time. The shortcut tuning method
that looks only at the initial change in ramp
rates and does not need to wait for a steady state
is the most reliable manual procedure for these
processes.

Always keep in mind the following six points

when approaching every application:

1. The step size in output or set point for a

tuning test should be at least five times the
noise band, valve dead band, and unmea-
sured load upsets.
 2. Unmeasured disturbances will deform the ing rule associated with the suggested method
response and stick-slip will cause a sus- selected. For example, a relay tuner or adaptive
tained saw tooth oscillation in the control- controller can be used and either Lambda or typ-
ler output of a self-regulating loop and in ical PID tuning rules chosen. For PID tuning,
the controlled variable of an integrating the normal Zeigler-Nichols factors applied to the
loop. ultimate period and ultimate gain are adjusted
per the general-purpose closed loop method.
3. When rounded oscillations appear in a Specifically, the controller gain is approximately
loop, they are usually the result of poor cut in half to provide more robustness and the
tuning or interaction of loops. To fin? th~ reset time is increased by a factor of 10 for inte-
culprit loop, put controllers one at a bme In grating and runaway responses. The rate time
manual until the oscillations stop, may also be reduced in some plug flow tempera-
ture loops to just equal the thermowelliag time.
4. Processes are nonlinear. The process gain, To mimic the shortcut tuning method, a pseudo
dead time, and time constant will change integrating process can be selected for the auto
with operating point and load so any. . tuner or adaptive controller and only the initial
attempt to get more than one or two sl~'l1fi- ramp used for calculating the tuning settings.
cant digits in tuning settings is an exerCise
in frustration. 6.1 Batch Control
5. Most loops where the process time constant Some of the more deceptively difficult types of
is larger than the dead time have too much batch control loops involve the simple heat up
reset action (too small of a reset time). or neutralization of a batch mixture where there
is no crystallization, evaporation, or reaction. If
6. Most loops where the dead time is larger. only a hot or cold heat transfer fluid (not both)
than the process time constant have too lIt- is available to bring a batch to temperature and
tle reset action (too large of a reset time). there is any reset action in the controller, the
batch will always end up beyond its set point.
While manual tuning methods are noted in this Similarly, if only an acid or base reagent (not
section, an auto tuner can be used and the tun- both) is available to bring a batch to a pH and

114 TUNING REQUIREMENTS FOR VARIOUS ApPLICATIONS

I
___-----A----
 there is any reset action in the controller, the Rule 43 - For simple single-
batch will always overshoot its set point. The ended temperature or pH control,
response appears to ramp. Particularly problem- a proportional plus derivative
atic is any residual heat left in the jacket or controller is needed to prevent
reagent in the dip tube when the control valve overshoot. The bias should be set
closes. to shut the control valve well before the CV
reaches set point.
In this guide the ability to go only in one direc-
tion with a manipulated variable will be called When the controller output drops to the point
"single-ended" control. This case occurs surpris- where seating friction and changes in the
ingly often. The solution is a proportional plus installed characteristic is significant, the control-
derivative controller. The bias should be set to ler output can be readily switched to pulse-width
provide zero controller output bef~re t~e con- modulation to get extra rangeability of the con-
trolled variable (CV) reaches set pomt smce the trol valve and eliminate the sensitivity, stick-slip,
CV will continue to coast for the duration of the and plugging problems associated with trying to
loop dead time after the valve is closed. Ho~- throttle near the seat. The conversion of a per-
ever, if the product is a validated phar~aceut1cal cent output to a percent pulse width can be
where a change in controller structure IS prob- done by a special output card or algorithm. The
lematic, most of the overshoot can be elimi- point for switching to pulsing is about 10% and
nated by setting the reset time a factor of ten or 20% for sliding stem and ball valves, respec-
more greater than would be calculated by a Zie- tively. If pulse width rather than pulse frequency
gler-Nichols or Lambda tuning rule. An open modulation is used, the valve gain remains con-
loop shortcut method is advantageous for these stant at its last value.
and other batch loops since there is often not
enough time or a steady state within the cycle Rule 44 - When the controller
time for a full tuning test. output drops low, use pulse-width
modulation to reduce plugging
and to improve the repeatability
of the end point.

116 TUNING REQUIREMENTS FOR VARIOUS ApPLICATIONS

~
6.2 Blending Rule 45 - Use Lambda tuning to
The accuracy and timing of feeds for a blend is get identical set point responses
often critical. The set point response for all the for the feed flow controllers. A
feed flow controllers should be identical. master composition loop may cause
Lambda tuning offers this capability. In some oscillations if it is too fast or has
cases there is an online analysis and a cascade too much dead time from violation of the cas-
control system where a master composition ~oop cade rule and destabilization, respectively.
manipulates the set point o~ the .feed loops III .
the proper ratio. The dead time III the compOSI- A model predictive controller can be set up and
tion loop heavily depends on the method and tuned to provide tighter composition control. If
location of the analysis, which sets the cycle the changes in feed composition or the desired
time and sample transportation delay. Analyzers composition of the blend are slow, a PI control-
that require process time or sample systems, cre- ler may suffice. Since the process gain varies
ate a severely dead time dominant master.lo?p. with the composition set point, gain scheduling
If fast inline sensors are used, such as Conolis, may be beneficial for large changes in the oper-
microwave, or nuclear magnetic resonance ating point.
meters, the process response of the master loop
is basically the secondary feed loops response set 6.3 Boilers
by Lambda tuning. This creates ~ unstable cas-
cade control system since the basICcascade rule The cascade control system where a drum level
of the master loop being 5 times sl~wer.than the controller manipulates the set point of a feedwa-
secondary loop is violated. The OSCIllatIOns can ter flow controller must handle the changes in
be suppressed by slowin~ ~own the mast~r loop steam use. Normally the level set point is con-
by detuning and the additIOn of a ~lter tune. A stant. Boiler drum level has an integrating and
feedforward of composition set POIllt changes noisy response with transients that are in the
can be added to the master loop output to opposite direction of the final response. While
improve its set point response. the cross-sectional area is relatively large because
it is a horizontal vessel, the ramp rate and inte-
grator gain is not exceptionally small because the

TUNING REQUIREMENTS FOR VARIOUS ApPLICATIONS 119

level range and residence time is smaller com- Rule 46 - Use mostly gain action
pared to a distillate or condensate receiver. and judiciously set the signal filter
for the drum level controller to
Bubbles and slosh cause measurement noise. An prevent shrink and swell from
increase in firing rate will increase the bubbles load upsets causing a level trip. If
per unit volume in the downcomers, wh~chwill the measurement is not noisy, rate action may
push liquid up into the drum and cause Its level help deal with the transients.
to swell. A decrease in firing rate will decrease
the bubbles per unit volume in the downcomers, When the controller increases or decreases the
which will cause the drum level to shrink. The feedwater flow, bubbles tend to collapse or
final level response after the steam generation expand, respectively, in the drum, which causes
sufficiently changes the drum liquid inventory is a response in the opposite direction of the fmal
in the opposite direction of the initial r~sponse response. This action would tend to counteract
to the change in firing demand. The shnnk and the shrink and swell from a firing rate change.
swell can be so severe that the boiler may trip on However, the inverse response for changes in
low or high level. The level controller needs to feedwater flow is usually quite minimal in mod-
be tuned with mostly proportional action and ern systems since the feedwater is preheated and
enough gain to prevent these trips. The reset the changes in bubbles in the drum causes a
time will cause overshoot when the inventory change in density that tends to reduce the
changes. While rate action could help deal ~th change seen by a differential head level sensor. If
shrink and swell, normally measurement nOlse the controller gain cannot be increased suffi-
makes this infeasible. A small filter or velocity ciently to deal with load changes, a transient
limiter on the level measurement can be used to feedforward signal can be added to the level con-
smooth out the signal enough to keep the feed- troller output that is proportional to the nega-
water valve from responding to level noise tive of the filtered derivative of the firing rate.
passed on as feedwater set point changes. How- The classical feedforward signal designed to
ever, a little bit of filter time goes a long way to enforce a mass balance that is proportional to
making an integrating loop less stable and slower the firing rate should be delayed so as not to
to recover from load upsets. interfere with transient correction.
 Rule 47 - If controller tuning is ler gain until an oscillation starts to appear after
not sufficient to prevent drum shrink and swell. A higher level controller gain
level trips, use a transient feedfor- will increase the size of the flow set point
ward to compensate shrink and changes, which in turn will help the flow loop
swell and a delayed classical feed- get through the valve dead band and resolution
forward to keep the mass in the drum con- faster.
stant. The transient feedforward and classical

fft ~g~~~~1~1;~:~1~~t
feedforward are proportional to the derivative
and magnitude of the firing rate, respectively.
The transient feedforward gain is negative
whereas the classical feedforward gain is positive.
feedwater valve. This means some
The feedwater flow controller should be tuned compensation of the feedwater valve nonlinear-
first with the Lambda method for an aggressive ity is desirable to enable a smaller than usual
response that will get through the dead band Lambda factor for the flow controller.
and resolution of the feedwater valve quickly.
The valve is the largest source of dead time in The oxygen controller normally manipulates a
the cascade control system. The process dead bias to the air flow measurement instead of the
time is less than a second. If the controller out- air flow set point to keep the cross limiting of
put can be characterized or the flow controlle~ the fuel and air controls intact. The set point
gain scheduled to compensate for the changes In may be optimized as a function of steam load.
slope of the installed characteristic of the valve, The oxygen response is dead time dominant
the Lambda factor can be reduced from 4 to 2. unless a large signal filter is used. It is a combina-
tion of dead time from the dead band and reso-
The level controller should be tuned with a lution of the air damper or control valve and the
closed loop method for load upsets. Changes in transportation delay of gas from the combustion
firing rate should be used during the tuning to zone to the oxygen sensor location. Some high
make sure it can deal with shrink and swell. temperature zirconium oxide sensors can be
Make sure the shrink and swell is not interpreted located in or near the combustion zone making
as a closed loop oscillation. Increase the control- the process transportation delay less than the

I
--..);------------
dead time from the damper or valve. A sample
6.4 Coils and Jackets
line to an analyzer house adds a considerable
transportation delay. Coil and jacket temperature loops are normally
se~ondary loops that must respond to the set
If the oxygen sensor is in the stack or analyzer pomt demands from a master temperature loop
house, dead time compensation or model pre- on a crystallizer, evaporator, fermentor, or reac-
dictive control (MPC) should be used with the tor. Th~se loops must also respond to distur-
dead time automatically adjusted to be inversely bances ~nthe coil inlet temperature or flow and
proportional to the load. The oxygen and air dependmg on the system, changes in the heat
flow controllers must be tuned for a smooth sta- transfer coefficient and area. These secondary
ble response, where robustness is more impor- co~trol.lers should be tuned with a closed loop
tant than performance. A larger than normal ?sclllatIOn method before the master controller
Lambda factor for PI control and move suppres- IStuned.
sion for MPC is advisable. The o~"ygencontrol-
ler should ignore the positive spikes in the :rn~ velocity is les~ and the dead time is greater
oxygen reading from the action of the cross lim- m.Jacket than a coll but otherwise the consider-
its for air to lead fuel for an increase in firing atIOns are similar. For the rest of this section
demand and air to lag fuel for a decrease in fir- coils will be discussed, but the conclusions are
ing demand. The oxygen controller must not the same for jackets.
chase noise at low load when the oxygen signal
gets noisy from poor mixing or at low oxygen L~w coil cool~t flow can lead to a high process
levels that are pushing the rangeability limit of gam and ~ead time that can cause instability. If
the sensor. at all pOSSIble,the coil flow should be kept con-
stant and the ~~keup flow manipulated to
Rule 49 - Tune the oxygen con- ch~nge the coll mlet temperature. For a recircu-
troller for a smooth and stable latIOn system, the return flow based on a pres-
response despite blips from cross ~ure balance should change to match the change
limits and noise from low loads. m makeup flow.

TUNING REQUIREMENTS FOR VARIOUS ApPLICATIONS 125

 Rule 50 - The coil or jacket cool- the recirculation system or at all for a once
ant flow must be constant to pre- through system.
vent oscillations for low cooling
demands. The makeup coolant When the heat transfer fluid is a vapor such as
flow is manipulated. steam, a secondary loop that controls pressure in
the coil would respond well to both changes in
If the secondary temperature loop controls the supply and heat transfer, if there is always a
coil outlet temperature for a batch reactor, it has demand for heat. This pressure loop has an inte-
an integrator response over part of the operating grating response and should be tuned accord-
range. For example, a tuning test with the .master ingly with a large reset time for minimal reset
loop in manual will show that a decrease III cool- action. Controlling pressure is also useful when
ant flow will cause an increase in coil outlet tem- you can't get a good flow measurement, which
perature but also an increase in reactor .' might be the case on a low pressure waste steam
temperature, which in turn causes an addlt10nal or heat recovery vapor stream.
increase in coil outlet temperature. Conse-
quently, the reset time of the coil outlet temper- Often both heating and cooling is needed and
ature loop should be increased by to be at least the coil loops have split ranged valves. The pro-
10 times the loop period. cess gain, dead time, and time constant is quite
different depending on which valve is throttled.
.. Rule 51- Tune the coil or jacket The switch over from one valve to another is a
...·f temperature controller first with a severe discontinuity. These loops often oscillate
closed loop method and increase around the split range point. The controller tun-
~. the reset time by a factor of 10 or ing settings must be scheduled based on the
more for outlet loops. valve throttled. Trim valves are used to reduce
the effect of stick-slip as one valve opens and
Coil inlet temperature loops respond faster to another closes. A tempered water system will
upsets in the chilled, cooling tower, or tempera- move most of the impact of these problems
ture water system. However, these loops do not away from the sight of the master loop.
see the upsets to the heat transfer coefficient or
area until after the transportation delay through
 Rule 52 - For split ranged valves, The controlled variable is often compressor dis-
schedule the coil or jacket con- charge pressure. Regardless of the type of manip-
troller's tuning settings based on ulated variable or compressor, Lambda tuning
the valve throttled and use trim should be used to provide a relatively slow and
(small) valves to suppress oscilla- smooth response to avoid upsetting users and
tions at the switch over point from heating to interacting with the surge control system. The
cooling. The transition from a trim to a coarse pressure set point may be slowly optimized
adjustment valve can be based on split ranged based on demand and efficiency to conserve
and simultaneous valve position control strat- energy.
egy.
Rule 53 - For compressor pres-
If the coolant flow is manipulated, the use of sure control, use Lambda tuning
heat transfer rate as the controlled variable to provide a gradual and smooth
improves the self regulation and linearity of the response to production rate
secondary loop. It also provides a record of the upsets and optimization of set
key variable of interest. After all, the master loop points.
temperature really depends on the heat transfer
rate, calculated as the product of the coolant Compressor surge control loops must be incred-
mass flow, mass heat capacity and temperature ibly fast to prevent surge. They must have fast
difference across the coils. The inlet temperature valves, scan times, and sensors and be tuned for
should be passed through a dead time bl~ck.so a fast response. However, even the best feedback
that a change in coil inlet temperature COlfiCldes loop may not be able to prevent surge or get a
with a change in coil outlet temperature. compressor out of surge. The precipitous drop
in flow occurs in less than 50 milliseconds and
the surge cycle period is only 2 to 4 seconds.
6.5 Compressors Consequently, feedforward and open loop back
Compressor capacity control loops are incredi- up signals are added to the surge controller out-
bly fast when motor speed is manipulated, but put to deal with upsets on a preprogrammed
generally sluggish when inlet damper~, guide basis. The feedforward signals are usually a sum-
vanes, or recirculation valves are manIpulated. mation of feed flows and a "kicker" that incre-

128 TUNING REQUIREMENTS FOR VARIOUS ApPLICATIONS

ments a surge valve open if the contr?ll~d
The run time between defrosting for continuous
variable such as suction flow, drops slgmficantly
operation and the cycle time for batch operation
below the surge controller's set point. The
depends on the quality of the inlet coil tempera-
"kicker" signal rapidly opens the surge valve ~d ture control.
gradually decays away to allow a smooth transI-
tion back to feedback control.

l1[~4t;~~£i¥~~~;~~
13[~~£~~~gE~~:~~~ low set point limit. Variability in
this loop adversely affects both capacity and
a

to production rate upsets. A feed- quality.

forward and kicker signal may need to be added
to the surge controller output.
The master temperature controller must be
tuned with rate action to deal with the heat
6.6 Crystallizers transfer lags and crystal growth rate kinetics. A
Crystallizers generally use a.cascade ~ontrol sys- shortcut open loop method is often the most
tem where the master loop IScrystallIzer temper- practical. For continuous operation the master
ature and the secondary loop is the coolant set point is fIxed, whereas for batch operation
temperature at the inlet to the coil or jacket. ~e the set point is programmed to provide the
coil inlet temperature must be kept from gettlI~g desired cooling and hence crystal growth rate
too cold to prevent the formation a.nd the rapId and size profile during the batch.
build up of small crystals on the COlI surfaces.
The frosting of the coil surface degrade~ the heat Rule 56 - For crystallizer master
transfer coefficient and proper crystal SlZeand temperature loops, use the short-
growth. A low temperature set point limit must cut open loop method and rate
be set and enforced. Lambda tuning is used to action to deal with thermal lags
provide a rapid but smooth response to set point and crystal growth kinetics.
changes and load upsets in the coolant system.

TUNING REQUIREMENTS FOR VARIOUS ApPLICATIONS 131

Some crystallizers are vacuumed controlled to
reduce the evaporation of solvent and the tem- go is usually determined by the smoothness of
the level measurement.
perature of the solution. Here a cascade control
system of temperature to pressure is used. The
pressure loop should b~ tune~ by a closed loop ~ Rule 57- When a distillate
method to provide an Immediate response by .. > ... ",,4Jt·· receiver level controller manipu-
!....••..•.•••.....••••.•
the use of mostly gain action before the temper- . :... .
lates reflux flow, use a closed loop
•.....
....( tuning method and a judiciously
ature loop is tuned.
set level measurement filter time
to maximize the controller gain.
6.7 Distillation Columns
The control scheme seen on many. columns The response of the column temperature con-
because it provides some inherent mternal reflux troller is generally incredibly slow and may take
control employs a level contr~ller on the over- hours or even days to oscillate or reach a new
head distillate receiver to manipulate the reflux steady state on large columns. Therefore, the
flow and the column temperature controller to shortcut open loop method is used if the closed
manipulate the distillate flow. However, ch.anges loop method takes too long. Rate action is essen-
in distillate flow only translate to change.sm tial to compensate for the interactive time con-
reflux flow that affect the column by actlOn ?f stants associated with the concentration
the level controller. Often the overhead receiver response of the trays. A feedforward of feed
is a horizontal vessel, which means lcu:gechanges changes added to the temperature controller
in distillate flow cause small changes m mea- . output is vital for improved control. If there are
sured level. The integrator gain for the process IS temperature controllers on the top and bottom
very low because the diameter ~sso large. ~or temperatures of the column to control the com-
this scheme to fulfill its potential, the rec~lver position in the distillate and bottoms, model
level controller must be tuned for a.very tight predictive control is essential to deal with the
response that requires controller gal~S m~ch interactions of 2-point composition control, to
higher than normall~ u~ed. Since g~m actlOn prevent flooding, and reduce steam use.
amplifies noise, the bmit on how high you can
Since tray temperature is an inference of compo-
sition and large concentration changes often

TUNING REQUIREMENTS FOR VARIOUS ApPLICATIONS 133

translate to small temperatures, precise tempera- For B~tch columns, the temperature set point is
ture measurement and control is important. a movmg target. Instead of temperature control,
Resistance temperature detectors (RTDs) and a level controller manipulates both the reflux
narrow temperature spans are use~ to ensure. and distillate flow in a ratio set by the batch
accuracy, resolution, and a good. slgnal-t?-nOlSe sequence. This affords the ability for the batch
ratio essential for the high rate orne settmgs ~peration to optimize the reflux without upset-
needed for tight controL tmg the level controller and the option to go to
total reflux.

13rI~~ff€;iffr~:~~~:~
interactive time constants of the
tray concentration response. Feedforward co~-
6.8 Dryers
Dryer temperature control loops suffer from
transportation delays and non-representative
product temperature measurements. While it is
trol, RTD sensors, and narrow temperature cah- desirable to measure the temperature of the final
product, the sensor isn't in good contact with
bration spans are criticaL For.2-J?oi~t
the flowing solids because of voids or buildup of
composition control and optlmlzauon, use
coatings on the thermowe1l. Consequently a gas
model predictive controL
temperature measurement is used. Sometimes an
There are many other difficulties encountered ~n inferen~ial moi~ture measurement (online prop-
the control of columns. For example, the manIp- erty estImator) ISconstructed based on an exit
ulation of steam to control sump level has an gas temperature subtracted from an inlet gas
inverse response from shrink and swell, and the temperature after it has passed through a dead
bottoms temperature controller v.:ill interact time. and filter block that match the delay and
with a sump level controller, particularly for a lag tIme of the dryer. Separate inlet and exit gas
temperature loops tend to fight each other. The
thermo-siphon reboiler. Chro?latog~aphs used
addition of a feedforward signal based on feed
for composition control pro~lde stau ca~ed .
response at best and an erratic and unreltable SIg- changes is also important. Model predictive con-
trol can sort out the interaction, deal with dead
nal at their worst. time, and provide better feedforward control. If

134 TUNING REQUIREMENTS FOR VARIOUS ApPLICATIONS

a viewing window can be kept sufficiently clean, direct moisture measurements can open the
an optical pyrometers may provide a more reli- door to more efficient operation.
able and accurate measurement of the product
temperature.
6.9 Evaporators
The product is usually much dryer than the The boiling point depends upon the composi-
specification because of deficiencies in the mea- tion and pressure of the contents. Often a tem-
surement and control system. By moving the perature loop manipulates the heat input, via a
dryer operating point closer to the high moisture coil or jacket steam flow or pressure controller
limit, moisture can be sold as product and and a pressure loop manipulates the vapor valve.
energy costs can be reduced. If the dryer mois- The boil up and vapor flow is proportional to
ture controller manipulates feed, the throughput the heat input. If an analyzer, such as a Coriolis
can be increased. The use of feedfOIward control d~nsity measurement, is installed in the product
and model predictive control, and online prop- dIscharge or recirculation line, a composition
erty estimators and microwave meters for mois- control loop can directly manipulate the heat
ture offer many opportunities for optimization input. The temperature and pressure measure-
of dryer operation that have been overlooked to ments become constraints. A model predictive
date. control can increase capacity of the evaporator
by m~imizing the feed flow without exceeding
".. Rule 59 - For dryer temperature the hIgh temperature and pressure limits.
•.•..
: ,\. controllers use a Lambda tuning
method with a small rate setting . '. . .".. Rule 60 - For evaporator temper-
~'. . equal to the thermowell time con- ature controllers, use a closed
stant. If a product temperature -= . loop tuning method with rate
measurement is used, the controller gain and ~ - action to compensate for thermal
rate action may need to be reduced to avoid lags. Use model predictive control
overreaction to an erratic measurement. For and a Coriolis meter for composition control
interactions and optimization, use model predic- and maximization of production rate.
tive control. Advanced control and inferential or

TUNING REQUIREMENTS FOR VARIOUS ApPLICATIONS 137

 6.10 Extruders variable response time and scan rates must be as
The temperature sensors for the zone tempera- fast as possible. If die opening is manipulated,
ture controllers are often measuring more the hydraulic actuators should be used. If the loop
extruder wall or jacket temperature than the cannot be made fast enough, a feedforward sig-
polymer temperature. The heat input from these nal based on extruder or sheet speed should be
zones act more as a warm blanket and is far added to the pressure controller output to help
exceeded in magnitude by the heat input from it deal with production rate changes.
the power to the main drive. The zone tempera-
ture controllers also interact and are adversely ~ Rule 62 - For extruder zone tern-
affected by split ranging. Lambda tuning is rec-
:Ii::Ji~~;iZ%~r~:
ommended.

....
Rule 61 - For extruder zone tem-
,,2~_:~
sure that upsets and nonlinearities

i3
...·...........i'.
....
perature control, use a Lambda
tuning method to provide a
smooth wall temperature. Make
sure that upsets and nonlinearities
in the manipulated variables do not affect the
in the manipulated variables do not affect the
wall temperature.

The extruder outlet pressure generally manipu-

lates the extruder speed. If another manipulated
wall temperature. variable exists, such as die opening, the speed is
freed up for polymer property control. An
The extruder outlet pressure generally manipu- online property estimator based on residence
lates the extruder speed. If another manipulated time and temperature rise may prove valuable
here.
variable exists, such as die opening, the speed is
freed up for polymer property control. An
online property estimator based on residence The extruder pressure loop must have a fast and
time and temperature rise may prove valuable smooth response. The sensor and manipulated
here. variable response time and scan rates must be as
fast as possible. If die opening is manipulated,
The extruder pressure loop must have a fast and hydraulic actuators should be used. If the loop
smooth response. The sensor and manipulated cannot be made fast enough, a feedforward sig-

138 TUNING REQUIREMENTS FOR VARIOUS ApPLICATIONS

TUNING REQUIREMENTS FOR VARIOUS ApPLICATIONS 139
nal based on extruder or sheet speed should be
nutrients, and reagents. In general, a temperature
added to the pressure controller output to help
water system is used to ensure the jacket temper-
it deal with production rate changes.
ature is moderated. Shocks to the cells from
localized extremes of any operating conditions
Rule 63 - For extruder pressure must be avoided.
controller, use a Lambda tuning
method, with fast hardware and
It is critical that the cells do not see any extremes
scan rates to provide a fast and
so the controllers are tuned for a gradual and
smooth response to load upsets. If
very slow response. Any tuning tests must be
this loop is not fast enough, a fe~d.for:w.ard
signal
short and involve very small excursions for short
of speed changes is usefu.l for mInlmlZI?g the
time intervals, which can be achieved by a short-
variability from changes In the produc~on rate. cut tuning method.
An online property estimator that mampulates
extruder speed may greatly improve product
quality. Rule 64 - For fermentor or
. ..J.......
bioreaetor control, use a shortcut
, tuning method with a greatly
6.11 Fermentors ~ reduced gain and decreased rate
Cell concentration, dissolved oxygen or carbon setting to provide a smooth and
dioxide, pH, and temperature are the important very slow response.
loops for fermentor and bioreactor co~trol. The
upsets are incredibly slow ~nd mostly Involve The loops interact and the split ranging of
the transitions from exceptIOnally low dem~ds manipulated variables changes the process
for the initial low cell concentration to the hIgh dynamics but the detuning and the slow nature
demands in the exponential phase of high cell of the upsets results in straight lines of the con-
growth and product formation r~tes and finally trolled variable throughout the batch. The effect
back to a low demand in the statIonary phase ~s of batch conditions instead show up in the
the cells mature. Split ranged manipul~te? var.t- trends of the controller outputs.
abIes are used to deal with extreme vanatIons In
demand for carbon dioxide or oxygen, cooling, The use of mass spectrometers for vapor analysis
and the calculation of oxygen uptake rates and

TUNING REQUIREMENTS FOR VARIOUS ApPLICATIONS 141

carbon dioxide production rates can be used to Rule 65 - For heat exchanger
create online property estimators. These estima- temperature controllers, a
tors along with strategies to optimize a proflle Lambda Tuning method with a
during a batch to improve yield or reduce cycle rate time equal to the thermowell
time will increase the need for better tuning and lag time is sufficient. However an
model predictive control. exchanger in a recirculation line for a vessel
should use a closed loop method with a 10
6.12 Heat Exchangers times larger reset time. Normally recirculation
temperature controllers are secondary loops
The response of a temperature control loop is whose set point is manipulated by a master reac-
dominated by a transportation delay and the tor temperature controller.
thermal lags of the thermowell. If coolant flow is
throttled, low velocities can cause a large dead
time and a high process gain and greater surface
fouling problems. If the coolant valve is The pH measurement offers an incredible sensi-
upstream of the exchanger, water can flash under tivity and rangeability that translate to excep-
some conditions. In general, the manipulation tional mixing, piping design, control valve
of a bypass flow instead of the coolant flow resolution, and controller tuning requirements.
greatly reduces the dead time and improves the A steep titration curve slope amplifies what may
controllability. In general, Lambda tuning is suf- normally be negligible amount of valve stick-slip
ficient unless the exchanger is in a recirculation and feed upsets. The nonlinearity can require
line of a vessel, which creates an integrating orders of magnitude change in controller gain
response. In this case, a closed loop method depending upon the operating point. The sched-
with the reset time set at least 10 times the loop uling of controller gain as a function of pH or
period and a rate setting equal to the thermowell the translation of the controlled variable from
lag is advisable. pH to percent reagent demand based on the
titration curve is highly recommended unless
the set point is on the relatively flat portion of
the titration curve.
 Rule 66 - For a pH set point on 6.14 Reactors
the steep portion of a titration
curve, schedule the controller The principal concern with reactors is the expo-
gain or translate the controlled nential dependence of the reaction rate on tem-
variable from pH to percent perature and the possible development of a
reagent demand based on the curve. runaway (open loop unstable) response. Plug
flow or catalyst bed reactors have a significant
Most pH loops on even well-mixed vessels dead time determined by the transportation
exhibit considerable dead time because of delay and an axial temperature profile. Back
reagent injection delays caused by small reagent mixed (stirred) reactors have a small dead time
flows. If the pH is driven from a knee to the set by the mixing delay and are designed to have
steeper portion of the titration curve, the accel- a uniform temperature throughout the mixture.
eration of the response can make the loop A closed loop tuning method and full rate
appear to lack self-regulation. The use of rate action is recommended for both. For severe exo-
action seems to help pH loops on vessels deal thermic reactors, such as those used in polYmer
with the nonlinearity of the titration curve and production, proportional plus derivative (PO)
pseudo runaway response. controllers are used because reset action is con-
sidered too risky. If a PIo controller is used on
Rule 67 - Use incredibly precise an exothermic reaction with a non-self regulat-
.. .t . reagent valves, specially designed ing response, the reset time should be increased
reagent injection systems, and a to be at least 10 times the loop period.
~ shortcut open loop tuning
method with considerable rate .. Rule 68 - For reactor tempera-
action for vc;ssel pH control. The measurement '. • Sture control, use a closed loop
will need tdbe filtered to keep fluctuations of method with full rate action. For
the controller output within the resolution of ~ . exothermic reactors, the reset time
the control valve. should be increased by a factor of
ten or more. For a severe runaway response, reset
action should not be used.
 When the product or byproduct is a vapor, a fied during testing is actually more of a problem
part of the vapor flow is often condensed and than a slower response, because MPC like the
returned (refluxed) to the reactor to improve the Smith Predictor is more oscillatory for an overes-
purity of the vapor stream. The exit temperature timate of the dead time. The existence of an off-
or heat removal rate of the overhead partial con- set also confuses MPC. If reset action cannot be
denser is controlled by manipulating the cooling used in sufficient amount to sufficiently elimi-
water flow or a small nitrogen flow to partially nate offsets in the set point response because the
blanket the condenser's heat transfer surface. RCAS loop response is not much faster than the
The condenser loop will interact with reactor MPC response, the offset can be added as a dis-
temperature and pressure loops by changing the turbance variable to the MPC.
reflux and vapor flow rates, respectively.
............
)' .. Rule 70 - Loops that operate in
Rule 69 - Partial condenser ".. the remote cascade (RCAS)
temperature loops should use a .- mode, should use a Lambda tun-
Lambda tuning method for a ~. - ing method to provide a consis-
smooth response with a rate tent and smooth set point
time set equal to the thermowell response with minimal offset. A fixed non-
oscillatory set point response with a negligible
offset is critical for minimizing model mismatch
in MPC.
6.15 Remote Cascade
Control loops that operate in a remote cascade 6.16 Sheets and Webs
(RCAS) mode where the set point is manipu-
lated by model predictive control (MPC) should Sheet production rate control is incredibly fast.
use a Lambda tuning method that provides a Each line roller speed must be properly ratioed
consistent and closed loop time constant. MPC to the manipulated master roller speed. Lambda
is sensitive to a dead time mismatch between its tuning and very fast scan rates are essential.
model and the actual response of its controlled Breaks in the sheet can cause an enormous waste
variables to the manipulated variables. A faster of raw materials and loss of production, particu-
response of an RCAS loop than what was identi- larly for large and fast sheet lines.

146 TUNING REQUIREMENTS FOR VARIOUS ApPLICATIONS

 Rule 71 - For sheet production Rule 72 - For PI control of sheet
rate control, use a Lambda tuning MD and CD thickness, use a
method for a smooth response Lambda tuning method and
with properly ratioed roller decoupled Smith Predictors with
speeds and a very fast scan rate. a dead time based on line speed.

Thickness control is a loop dominated by the

transportation delay from the die to the density
or thickness gage. The dead time in a Smith Pre-
dictor or MPC must be updated based on sheet
speed. The control of the thickness along the
length of the sheet is called machine direction or
MD control. The control of thickness across the
sheet is called cross direction or CD control. For
MD control, the manipulated variable is a take-
up roller speed that is ratioed to the master roller
speed used for production rate control. For CD
control, the manipulated variables are a series of
die bolts with actuators or heaters. To date, MD
and CD control has been predominantly done
by decoupled Smith Predictors. Exceptionally
large and fast matrix algorithms has enabled the
migration ofMD control and some CD control
to MPC for optimization. However, the con-
trolled variables for wide sheets that have a 100
or more CD controlled and manipulated vari-
ables still require some consolidation of vari-
ables to fit within today's MPC.

TUNING REQUIREMENTS FOR VARIOUS ApPLICATIONS 149

7.0-Adaptive Control

7.1 Learning the Terrain

The next generation of adaptive controllers will
identify the process dynamics and allow the user
to pick the most appropriate tuning rule for their
process. By identifying and remembering the
process terrain, the controller does not have to
rehash old information and can concentrate on
changes. Knowledge of the changes in the open
loop gain, dead time, and time constant enables
recognition of changes in raw materials, equip-
ment, conditions, and the response of instru-
mentation and control valves that are important
for plant performance. For example, a decrease
in valve gain can be caused by a partially
plugged filter and an increase in loop dead time
can be caused by a fouled thermowell.

Rule 73 - Use knowledge of the

... changes in the process dynamics
_. to diagnose equipment and instru-
~- mentation problems. The changes
in process gain often reflect changes
in the composition, pressure, and temperature of
streams, whereas changes in the process dead
time often reflect changes in the fouling of heat
transfer and sensor surfaces.
7.2 Watching but Not Waiting
If the process dynamics from previous excur-
sions in different operating regions of the actual
plant or a virtual plant (dynamic process model)
are stored as a function of key process variable or
manipulated variable, the controller tuning can
be scheduled. Figure 23 shows 4 different pH
regions with vastly different process gains that
reflect the nonsymmetrical shape of the titration
curve for a mixture of acetic acid and ammonia
with a dash of sulfuric acid and the absorption
of carbon dioxide from the air. This allows the
adaptive pH controller to take preemptive action
when it moves to a new operating point. Previ-
ously, adaptive controllers had to see step
changes or oscillations even though the topogra-
phy had not changed. These controllers were
always playing catch-up.

Rule 74 - Schedule the control-

ler tuning settings based on
knowledge gained from previous
excursions into different operat-
ing regions of an actual or virtual
plant.
7.3 Shifting into High Gear
Most processes are not operating at the most ~
lIfl
optimum set point primarily because of a lack of ;J
•••
process knowledge and excessive process vari- =~
bO
ability. The set point chosen by operators are
based more on tradition and opinions rather ~
~
than an understanding of more efficient operat- y

ing points. A zoom-in on the titration curve in ""==

the control region in Figure 24 shows the oppor- ~
.•..e
tunity for the neutralization of an acidic influent
by lowering the set point closer to the low pH
limit. If the influent becomes basic, the set point
-t
~
would be switched to be closer to the high pH U
=e
limit. In this case, stick-slip and feed upsets ::
Q,
(oscillations on the X axis) have less impact at ~
~
the more optimum set point. '':=Q,
tt
.. Rule 75 - Use process knowl- ""=
<:
" edge and an adaptive controller to ;
move the process to a more effi- c.S••
~ cIent operatIng POIOt. .•C J:
Q.

~
i.
Q,

0
I
'l:t
N
~

..'"'-
~
7.4 Back to the Future
1
i
$
.~

the adaptive controller for a higher level of con-

This new generation of an adaptive controller i
,!
trol by a better knowledge of the topology.
allows all PID loops to run in the adaptive mode Online process performance indicators with eco-
and provides process model parameters that are nomic factors, such as the cost of excess
saved in a data historian and analyzed for reagents, reactants, or energy, can provide the
changes in the plant, sensors, and valves. The knowledge, motivation, and justification to
information on changes in the process model invest in the use of new technologies to find and
may be directly used to monitor loop perfor- exploit more efficient operating points. Figure
mance and to provide more intelligent diagnos- 25 shows how an online indication of excess
tics. The models can provide the dynamics for reagent cost was greatly reduced by the use of an
simulations and identify candidates for feedfor- adaptive pH controller to get to a more efficient
ward control and advanced control techniques. set point.
For example, loops dominated by a dead time or
exhibiting disturbance models for multiple vari-
ables, are prime candidates for model predictive •...•....
~!:c::di~~~r~~~';r~:~~rilie
control. Feedforward models of compositions knowledge, motivation, and justi-
could be used for dynamic online property esti-
mators and loops dominated by a single large
~ - fication to install and keep
improved control systems in their
time constant could benefit from fuzzy logic highest mode.
control. The dynamic process models in general
If you understand and practice everything in this
can be used to create or adapt real-time simula-
pocket guide, you might become the next CEO
tions for prototyping new control strategies,
of your company. If this is not realistic, maybe
exploring "what-if" scenarios, and training oper-
you will be given stock options for every loop
ators. Process gains that decrease or time con-
that is tuned. If this is a pipe dream, the
stants that increase with feed totals are ripe for improvement in loop performance might lead to
real-time optimization of the run time between a promotion. Well, maybe you will get a gift cer-
defrosting or cleaning and catalyst reactivation tificate to Burger King.
or replacement. The beauty of this route is the
models and tuning settings are available from Seriously, I wish you all the best of luck in tun-
ing loops.
Appendix A-Technical
Tenns in Process Control
that Are Used Interchange-
ably
Attenuation - Filtering - Smoothing

Delay Time - Dead Time*

Derivative Action - Rate Action**

Derivative Time - Rate Time**

Digital Filter - Measurement Filter - Process

Variable Filter - Signal Filter

Integral Action - Reset Action **

Integral Time - Reset Time**

Lag Time - Time Constant*

Process Variable Gain - Process Gain*

 Appendix B-For Math
Scan Time - Cycle Time - Execution Time Lovers Only
Steady State Gain - Static Gain *
The simplified Lambda tuning equation f~r .the
controller gain can be derived from t~~ ongmal
System Dead Time - Total Loop Dead Time*
equations for Lambda tuning. The ongmal equa-
tions are useful for designing loops to have the
Upset - Disturbance - Load Change
same closed loop time constant for blending
operations and model predictive .control (M~C),
particularly when there are. multlpl~ producnon
lines. In these cases, EquatIon B-1 IS solved for
* Dynamic Model Parameters
the Lambda factor (4) that gives the desired
closed loop time constant ("te).
** Controller Tuning Parameters

Original equations:

"tc = At- * "to (B-1)

Tj = T98/ 4 (B-2)

Kc = Tj / [ IS> * ("tc + "tdo )] (B-3)

Time to 98% of final response is 4 time con-

stants plus the dead time:
Substituting Equation B-4 into Equation B-2 ICe = 1/[Ka*41 (B-9)
shows the dependence of the integral time on
the open loop time constant and the observed Substituting the definition of the open loop gain
dead time: (Kc = ~%PV / ~%CO) into B-9, we end up with
the simplified equation for the controller gain
shown in section 2.5:

Substituting Equations B-1 for the closed loop ICe = (~%CO 1 ~%PV) 1 4 (B-1 0)
time constant ('tJ into the denominator and B-5
for the open loop time constant corrected for
dead time into the numerator of B-3:
~%CO= change in controller output (%)
~%PV = change in process variable (%)
I\: = controller gain setting
Factoring out the Lambda factor in the denomi-
Kc = open loop gain (~%PV / ~%CO)
nator:
Ar = Lambda tuning factor
Tj = controller integral time setting (see!
repeat)
Substituting the La.mbda fac~or for robustI.less 'tc = closed loop time constant (e.g., set
(Ar = 4) for the deVIsor Of'td m the denommator point response time constant) (see)
otB-7: 'tdo = observed dead time (see)
'to = open loop time constant (e.g., largest
Kc = ('to + 'tdcl4)/[Ko * 4* ('to + 'tdo 14)] (B-8)
process time constant) (see)

Canceling out the common term

Note that ~%PV is the same as ~CV since the
controlled variable is the process variable
expressed as a percent of the measurement span
corresponding to the controller scale.
Appendix e-An Integral
Part ofTIIDing
The integral time setting presented at the end of
Chapter 2 is important for the secondary loops
of cascade control systems and for integrating
and runaway processes in general. For primary or
single loops on self-regulating processes, the fac-
tor for the closed loop tuning method applied to
the loop period to get the best integral time set-
ting varies from about 1/8 to 1.0 for self-regulat-
ing processes, which corresponds to a pure dead
time and lag response, respectively. Equation
C-l has been shown to be effective at estimating
an integral factor of 1/4 to 1 for industrial systems
where pure dead time processes are rare.
..,
~ = 1I{[(4 * 'td,ffo) - 1)]- * 3 + I} (C-l)

Ti = ~ * To (C-2)

If'td < To / 3 then (otherwise ~ = 0)

~ = 0.1 * kj
(C-4)
For the shortcut method, a process gain of one is = derivative time factor applied to total
assumed so that the process time constant is the dead time in shortcut method
inverse of the pseudo integrator gain. The short-
cut dead time factor is then the closed loop
= integral time factor applied to loop
period factor multiplied by the loop period period in closed loop method
divided by the dead time. = derivative time factor applied to loop
period in closed loop method
(C-S) = change in the process variable within
the 1st dead time interval (e.u.)
(C-6) = change in the process variable within
the 2nd dead time interval (e.u.)
The following equations have been used to = change in the process variable within
approximate the loop period in various applica- the 3rd dead time interval (e.u.)
tions for self-regulating processes: = controller integral (reset) time setting
(see/repeat)
To"" 1.4 * {[(l_e-oetd/'to)/(l_e-(td/'to)] + I} * "Cd = controller derivative (rate) time setting
(C-7) (sec)
= quarter amplitude loop period (sec)
To = 1.4 * ([(M'V1 + ~PV2 + ~3) /~PVd + I} * 'td = total loop dead time (sec)
(C-8) = open loop time constant (e.g., largest
process time constant) (see)
To = 1.4 * {I + l('td /('td + 'to)] 0.65 + I} * 'td
(C-9)

Cj = integral time factor applied to total

dead time in shortcut method
Appendix D-Closed Loop
Time Constant
The time constant used in this guide to charac-
terize a process response is an open loop time
constant, which is the time constant for the
response of the controlled variable to a change
in controller output with the controller in man-
ual. It is increasingly popular to specify a closed
loop time constant, which is the response of the
controlled variable to a change in controller set
point with the PIo algorithm active (Auto, Cas,
or Rcas modes). For loops whose set point is
manipulated by a batch, blend, model predictive
control, or ratio control system, a closed loop
time constant is an effective way to specify a
consistent set point response that is essential for
the performance of these systems.

Lambda tuning provides a closed loop time con-

stant (LJ that is approximately equal to the open
loop time constant (La) multiplied by the
Lambda tuning factor (At-).

(0-1)

The closed loop time constant (LJ can be esti-

mated for any tuning method as the integral
time (TJ (i.e., reset time) setting multiplied by
the product of the controller gain (KJ and the
Index
dimensionless open loop gain (Kc) defIned in analog/digital converter
A
Section 1.6. abrupt responses 19, 27 (NO) 27, 29, 63, 76
accuracy analyzer
auto tuner 1 cycle time 27
measurement (see attenuation (see filter)
measurement auto mode 6
resolution) auto tuner
tuning 1 accuracy 1
action identification of dead
conuol2-6 band 85
direct 2-6 integrator gain 40
Eail-open 2-6 pretest 87
gain 7,14,16,22,96-97, relay method 85-86
101-102, 105 Jutomatic mode I, 6
on-otT99, 101, 107 B
process 2-6 back-olcu1.lte
nte 8,14,38,52,95,97, function blocks 4, 7
101-102,105 backlash (see dead band)
~t7, 14, 16,20,22,24, ball valve 66, 71
51-52,85,88-89,94, batch control tuning 8, 16,
96-97 44,115
reverse 3-6 blending 118
valve 2-6 (see also pipeline Jnd
actual valve position 57 static mixer)
actuator 66, 71 boilers 119
adaptive control 151 boiling point 137
advanced conuol tuning 8, bump 43, 51, 53, 60
16,44 butterfly valve 12,36,66, 71
algorithm
error-squared 20, 52 C
calibration 64
PID 9, 47
calibration span 11, 27, 37
update time 105, 110-111
cas (cascade) mode 6
aliasing signal 105
cascade control 3, 8-9, 16
analog output block 4
 cascade control tuning44,48, ofl3et 12,24,96,109 valve 2,19-20,29,33, dynamic classes
95-98 oscillations 16,21-25,34, 47-48,51,53,55,60, valve 66, 69
characteristic 48-55,58,70,85,88 64,71,73 E
valve 12,22,31,36-37, overshoot 8, 14,24-25, dead time error
71,73,95 101 of valve 47, 53, 55-56, 58, control 8, 28
characterization performance 28, 33, 92, 60,71,73 error control 7-8, 11,28
signal 71, 73, 94 95, 111 dead time or time delay error~quared algorithm 20.
chatter 27-28 residence time 35, 37 response 30-34,53-57, 52
closed loop 6-7, 44, 47-48, resonance (loops) 99, 101 88-90,92,95,97 evaporator control tuning 14,
51-57,58 set point 8, 14, 16-18,20, dead time to time constant 21,96
coated 44, 78, 88 23 ratio 16,35 evaporators 137
coating of sensor 63-64, 85 split-ranged 14 dead-time compensator execution time (see scan
coils 125 stability 22, 24 tuning 88-92, 108 time)
inlet temperature 126 upset (see also dead-time dominant exothermic 145
column control tuning 14, disturbance) 30, 52-53, response 16, 19,35-36, extruder outlet pressure 139
19,21-22,24,36,53,92 55,60,92,95-96,99, 52,58,101,108 extruders 138
column control valve 70 101,105,107 delay
compression 105, 110-111 variability 1-2, 16, 19,27, from transportation 33, F
compressor 128 107 88-89 faceplate
control tuning 37, 39 controlled variable 11, 37, 90 delay time (see dead time) PID 14
control valve 66, 70 controller output derivative f.til-<>pen
surge control 129 function blocks 4, 12 action (see rate Jction) valve 4-6
concentration control tuning PID 12,40,51,55,57,60 mode 8,10 falter or hesitate response 16,
27,29,38,85,96 conveyor control tuning 19, time (see rate time) 101
control 24,58 desuperhc:ater control tuning teedback control 2, 33, 88
action 2-6 crystallizer control tuning 14, 19,36,38,58 fc:c:dforwardcontrol 52, 92-
bump 43, 51, 53, 60 21 digital filter (see 94,96
cascade 3, 6, 8-9,16,19 crystallizers 130 measurement filter) fc:c:dwaterHow controller 122
disturbance (see also current-to-pneumatic digital positioner 64, 71 termentors 140
upset) 30, 88, 92, 95 transducer (liP) 4 direct action 2-6 tieldbus function blocks 4, 9,
error 7-8, 11, 28 cycle time (see scan time) direct digital control (DOC) 98
feedback 2,33,88 cycle time analyzer 27 7 fieldbus mode 6-7
feedforward 52, 92-94, 96 distillation 132 filter 8, 12, 19,27,51,55,60,
interaction 16,20,44,52- (sc:c:also column control) 94-95, 97
0 flow (feed) ratio 36-40
53,58,62,95,97,99- dead band 19,47,53,56,64, disturbance control (see also
100 upset) 30, 88, 92, 95 flow control tuning 19, 38,
73-74
limit cycle 55 drift sensor 63-64 52, 85, 89, 95
material balance 21 dryers 135

172 INDEX INDEX 173

 fluctuations or waves integrator gain loop initialization manual
pressure 99-100 for auto tuner 40 closed 6-7, 44, 47-48, 55, (IMan) 7
fouled 88 (see also coated) integrator response 39-41 58 integral 7, 10-11
function blocks interaction 95, 97 master (also known as local (see auto) 6
back~lculate 4, 7 control 16, 20, 44, 52-53, primary) 8,56 local override (La) 6
controller output 12 58,62,99-100 open 6,47, 53, 58 manual (Man) 6
fiddbus 4, 9, 98 inverse response 27, 52-53, primary (also known as operational 6
PID4 61,90,92,94 master) 8, 56 output tracking 6
s«ondary (also known as proportional 7, 9-12
G J
gain action 7, 14, 16,22,96- jackets 125 slave)8,48,95-98 remote (see cascade) 6
97,101-102,105 slave (also known as remote cascade (RCas) 7
K secondary) 8, 48, 95-98 remote output (ROut) 7
gain tuning 7-11,19-24,35, -kicker" algorithm 129·130
47-48,51-52,55,57,60- troubleshooting 99-111 remote set point (RSP) 6
61,102,109-110 L supervisory 7
M
general-purpose method lag model predictive conuol
maintenance
tuning 44-48 thermowell 29 (MPC) 28, 124
of valve 64, 71-72
time (see also time manipulated variable 3, 9, II, N
H constant)
heat exchanger conuol 22,28,31,36,39-40 neutralizers 143
Lambda factor 122 manual (Man) mode 6 noise signal 19-20, 27, 30,42,
tuning 19, 36, 38 Lambda tuning 35, 38,57-
heat exchangers 142 master loop (also known as 44,51,55,60,63,70
62,96-97 pnrnnary)8, 56, 95-98 noisy measurements 81-84
heat transfer 127 level
historian master temperature nonlinearity response I, 36,
conuol valve 63, 70-71 controller 131 47,58,94-95
compression 105 controller 122 material balance control 21
update time 105 switches 99, 105, 107 a
measurement filter 29,55,60, offset control 12,24,96, 109
I tuning 19-21,24-25,29, 133
identification of dead band 44,51-52 on-off action 99,101, 107
measurement resolution 27- open loop 6,47,53,58
85 limit cycle 55 28, 110-111
increase-to-close valve 3-6 limits open-loop gain response 34-
methods
inferential moisture output 51, 97-98 42,57,60-61
tuning I, 16 operational mode 6
measurement 135·136 set point range 97-98 mode
initialization manual (IMan) set point velociry 8, 19, oscillations 16,21,24-25,34,
auto 6 48-55,58,70,85,88
mode 7 52 automatic I, 6
integral action (see reset load change (see disturbance) control and response 16
cascade (Cas) 6 output limits 51, 97-98
action) local mode 6 derivative 8, 10
integral mode 7,10-11 local override mode 6 output tracking 6
direct digital control overshoot control 8, 14,24-
integrating response 47, 108
(DOC) 7 25,101
fieldbus 6-7

174 INDEX INDEX 175

 oxygen controller 123 waves 99, 101 ratio chatter 28
oxygen sensor 124 pretest dead time to time dead time or time delay
p auto tuner 87 constant 16,35, 108 53-57,88-90,92,95,
primary loop (also known as Bow (feed) 36-40 97
patterns of response 92
master loop) 8, 56, 95-98 rate time to dead time 52 dead-time dominant 16,
performance
control and tuning 28, process rc:sc:ttime to oscillation 18-19,35-36,52,58,
action 2-6 period 24, 48, 52 101, 108
33,92,95, 111
dynamics 15.2 signal-to-noise 63 falter or hesitate 16, 101
pH
gain 36-39, 89-90,95, reaetorcontrol tuning 14,16, integrating 47, 108
control tuning 27, 29, 31,
114, 119, 125, 127, 21-22,44,48,51,96,109 integrator 39-41
36,55
control valve 70, 88
142,151,156 re3etors 145 inverse 27,52-53,61,90,
variable 2-3, 8-9, II, 14, reagent demand control 143- 92,94
measurement 143
16,19,27,31,36,40, 144 nonlinearity 1,36,47,58,
PID
algorithm 9, 47 51,57-58,60,88,90 rday method 94-95
varia ble filter (see auto tuner 85-86 open-loop gain 57, 60-61
controller output 12,40,
51,55,57,60 measurement tilter) remote oscillations 70
variable gain (see process cascade mode 7, 146 patterns 92
faceplate 14
gain) mode (see cascade) 6 quarter-amplitude 21- 22
function blocks 4
proponlonal output (ROut) mode 7 fUn3Way47-48, 108
scan time 11,28-29,45-
action (see gain) set point (RSP) 6, 98 self-regulating 40
46,53,63,85,95,97,
band tuning 9, II, 20 reset sensor 63
107, 110
mode 7, 9-12 action 7, 14, 16,20,22, set point 44, 48, 51-52,
pipeline control tuning 19,
24,27,38,58,89 pulp and paper control 24,51-52,85,88-89, 55,60
tuning 20 94, 96-97 smooth 19-20, 35, 88, 96
positioner
digital 64, 71 pump control tuning 37, 39 time to oscillation period staircase 25, 101
pneumatic 64, 70 Q ratio 24, 48, 52 steady-state gain 90
Srtl3rt (see digital quarter-amplitude response tuning 10-11,14,22-35, time constant 8, 16,20,
positioner) 21-22 47 27,37,40,89-90,95,
valve 4, 64-71 residence 97
R time control 35, 37 time to steady state 58-
power spectrum signal 105- ramp rate 39-41, 53-55,107,
106 temperature detectors 63 62,66-69
109-110 resolution trajectory 14,29
pressure rate
fluctuations or waves 99- measurement 27-28, reversal
action 8, 14,27,38,52, 110-111 of signal 3-6
100 95,97,101-102,105
control tuning 16,24,35- valve 27-29, 64, 71 rotary valve 66, 70-71
time to dead time ratio 52 runaway response 47-48, 108
37,44,48,51,85,96, response
tuning 8, 10,25-27,47,
100 abrupt 19, 27 response
58 analyzers 27 runaway 22
control valve 66, 71

176 INDEX INDEX 177

s shrink and swell 12Q-122,134 split range valve 70 trajectory of response 14,29
scan time 28-29, 45-46, 53 signal split-ranged control 14 transmitter
PIn 11,63,85,95,97, aliasing 105 stability 22, 24 smart 63-64
107, 110 analog to digital staircase response 25, 101 transportation delay 33, 88-
scheduled tuning 143-144, converter (NO) 27,29, startup control tuning 16 89
152 63 static gain (see steadY-5tate trend recording
secondary (also known as current-to-pneumatic gains) compression 105, III
slave) 95-98 transducer (liP) 4 static mixer control tuning update time 105, 111
secondary loop (also known filter 8,12,19,27,51,55, 19,36,38 troubleshooting loop 99-111
as slave) 8, 48, 95-98 60,94-95, 97 (see also steadY-5tate gains 3Q-42, 90 tunlOg
self-regulating response 40 measurement filter) steam traps 99, 101 accuracy 1
sensitivity (see resolution) noise 19-20, 27, 30, 42, step size 43, 53, 58, 66, 69, advanced control 8, 16,
of valve (see valve 44,51,55,60,63,70 100,110-111 44
resolution) power spectrum 105-106 stick-51ip 74-77 batch control 8, 16, ~
to measurement (see signal characterization 71,73, valve 31, 64, 102,107 cascade control 44, 48,
measurement 94 sticktion (see resolution) 95-98
resolution) signal reversal 3-6 Stroking time (see valve closed loop method 51-
sensor signal-to-noise ratio 63 response time) 57
coating 63-64, 85 single-cnded control 116 supervisory comeol tuning 7 column control 14, 19,
drift 63-64 sIZe supervisory mode 7 21-22,24,36,53,92
response 63 step 43, 53, 58, 66, 69, system dead time (see total compressor control 37,
sequence control tuning 19, 100,110-111 loop dead time) 39
44 slave loop (also known as concentration control 27,
set point secondary) 8,48, 70,95-
T
temperature 29,38,85,96
control 8, 14,16-18,20, 98 conveyor control 19,24,
23 control tuning 14,24,27-
sliding stem (globe) valve 71 58
range limits 97-98 smart positioner (see digital
29,35,38,48,52
control valve 70-71 crystallizer control 14, 21
response 44, 48, 51-52, positioner) dead-time compensator
55,60 smart transmitter 63-64 detectors 63
loop controls 126 88-92, 108
velocity 8, 19 smooth response 19-20,27,
thermowelllag 29 desuperheater 36
velocity limits 52 35,88,96
time constant response 8,16, desuperheater control 19,
settings for tuning 43,45,51, smoothing (see filter)
20,27,30-37,40,89-90, 38,58
53 span evaporator control 14,21,
shaft of valve 66, 68, 71, 73 calibration 11,27,37 95,97
time delay (see dead time) 96
sheet (web) control tuning specification flow control 19, 38, 52,
19-20,24,58 valve 66 time to steady state response
58-62 85,89,95
sheets 147 spinning (fibers) control
shortcut tuning 85 tuning 19 total loop dead time 42, 62,
88
gaUl35,47-48,51-52,55, static mixer control 19, rnaUltenance 64, 71-72 Z
57,60-61,102,109- 36,38 pH control 70, 88 Zic:g1c:r-Nichols
110 supervisory control 7 positioner 64-71 tuning 21, 48
general-purpose method tempc:l'2tUIc:control 14, pressure control 66, 71
44-48 24,27-29,35,38,48, resolution 27, 28, 29,64,
hat exchanger control 52 71
19,36,38 web control 19,24,58 response time 66-69
Lambda 35,38,57-62, window of allowable rotary 66, 70-71
96-97 gains 22, 47 shaft design 66, 68.71.73
level 19-21, 24-25, 29, Zic:gler-Nichols 21, 48 size 11
44,51-52 tuning methods 1, 16, 47-62 sliding stem (globe) 71
pH control 27, 29,31,36, U specification 66
55 update time of algorithm 105, split range 70
pipeline control 19, 24, 110-111 stick-slip 31, 64,102,107
27,38,58,89 upset control (see also temperature control 70-
pressure control 16, 24, disturbance) 52-53. 55, 71
35-37,44,48,51,85, 60,92,95-96.99,101, variability 1-2.16,19,27,
96, 100 105, 107 107
proportional band 9, 11, variable
20 V controlled 11, 37, 90
pulp and paper control 20 valve manipulated 3.9,11,22,
pump control 37, 39 actual position 57 28,31,36,39-40
rate 8, 10,47,58 actuator 66. 71 process 2-3, 8-9, 11, 14,
reactor control 14, 16, ball 66, 71 16,19,27,31,36,40,
21-22,44,48,51,96, butterfly 12,36,66, 71 51,57-58,60,88,90
109 characteristic 12,22,31, variable speed drive (VSD) 3,
requirements 113-115 36-37, 71, 73, 95 36
r~etlo-ll, 14,22-35,47 column control 70 velocity 52, 125
sequence control 19, 44 compressor control 66,
70 W
settings 43, 45, 51, 53
dead band 2, 19-20,29, waves
sheet (web) control 19-
33,47,48,51,53,55, pressure 99, 101
20,24,58
60,64,71,7.3 web control tuning 19,24,58
shoncut 85
dead time 47, 53, 55-56, webs 147
slave loop 8, 48, 70, 95-
58,60,71,73 window of allowable gains
98 tuning 22, 47,109-110
sPUlning (fibers) control dynamic classes 66, 69
19 fail-open 4-6
startup control 16 level control 63, 70-71